System and method for automating exploratory data analysis based on variable attributes

Abstract

The present invention relates to a data analysis automation system, and more particularly to a variable attribute-based exploratory data analysis automation system that can automate data analysis by taking into account variable attributes so that an algorithm can be adaptively selected for variously generated variables to perform data analysis.

Description

[Technical field]

[0001] The present invention relates to a data analysis automation system, and more particularly to a variable attribute-based exploratory data analysis automation system that can automate data analysis by taking into account variable attributes so that an algorithm can be adaptively selected for variously generated variables to perform data analysis. [Background technology]

[0002] In clinical fields, software equipped with various statistical analysis algorithms is frequently used to look for factors that cause or are related to the development of diseases, and to analyze the effectiveness of newly developed drugs and treatments.

[0003] Clinical researchers (doctors, nurses, pharmacists) must prepare clinical papers in order to produce research results, and clinical statistical analysis is necessary when preparing such clinical papers. However, it is extremely difficult for individual clinical researchers who have insufficient knowledge of statistical analysis to perform all of this clinical statistical analysis without the help of experts.

[0004] In particular, the terminology used varies depending on the medical department, such as gastroenterology / cardiology, neurosurgery, orthopedics, and is composed of highly specialized words, making it extremely difficult for an individual to perform statistical analysis, which requires extensive study, without the help of an expert. Furthermore, since the experience of the researcher who produced the data is very important for data analysis, even a statistician without clinical knowledge cannot properly perform clinical statistical analysis alone without the help of a clinical researcher.

[0005] Thus, there is a need for software programs that can easily perform statistical analyses suitable for clinical researchers who are familiar with the characteristics of the data and the study objectives.

[0006] Conventional statistical analysis software is littered with statistical algorithms and parameters, making it difficult to use and analyze the analysis results. The process of manually extracting and editing the results is time-consuming and prone to editing errors.

[0007] Therefore, it can be said that it is extremely difficult for a non-statistician to use conventional statistical analysis software to correctly calculate the analytical results that the researcher desires.

[0008] The inventor of the present application invented and proposed a method for automatically searching for variable association relationships and calculating a dynamic result report using the same, which is disclosed in Korean Patent Application No. 2011-0104734.

[0009] The method for automatically searching for variable correlations and calculating a dynamic result report using the same grasps the characteristics of many clinical and epidemiological variables used in medical statistics, automatically classifies them into the most commonly used types, and when there are many variables to be analyzed and statistical algorithms to be applied, automatically specifies correlations between all variables, thereby providing a comprehensive result report that allows a user to grasp at a glance the numerous correlations analyzed between clinical and epidemiological variables.

[0010] Such prior inventions are based on the idea that, since most analytical parameters commonly used in statistical analysis are fixed, and even if the analysis is complicated and performed by a statistical expert, if the process is analyzed in detail, it can be converted into a process of selecting two or more conditions that can be propositioned through several stages, and therefore a statistical algorithm can be automatically selected according to the characteristics of the variables.

[0011] The technical feature of the present invention is that it grasps the characteristics of numerous clinical and epidemiological variables used in medical statistics, automatically classifies them into the most commonly used types, and when there are many variables to be analyzed and many statistical algorithms to be applied, automatically specifies the correlation between all the variables one by one.

[0012] Such a conventional method for automatically searching for variable associations has a technical feature in that it separates types of clinical and epidemiological variables in consideration of their characteristics according to a process defined in a statistical algorithm, and determines an automatically applied statistical algorithm accordingly, so that the statistical algorithm is automatically applied according to the variable characteristics, thereby allowing even a user who is not familiar with statistical analysis methods or detailed statistical algorithms to easily use a statistical analysis program.

[0013] In recent years, clinical processes and fields have become more diverse and specialized, and in clinical settings such as hospitals at the clinical stage, data on tens to hundreds of variables is produced depending on the field. This makes it difficult to specify the relationships between all of the variables, and limits the ability of clinical researchers, who are mostly non-experts in the field of statistics, to select appropriate statistical algorithms. [Prior art documents] [Patent documents]

[0014] [Patent Document 1] Korean Patent Publication No. 2011-0104734 Summary of the Invention [Problem to be solved by the invention]

[0015] The present invention proposes a data analysis automation system as an extension of the above-mentioned conventional "automatic variable association search and dynamic result report calculation method using the same" and proposes a data analysis automation system that is adaptive to variables that are generated in a more diverse manner. It is an object of the present invention to provide an exploratory data analysis automation system based on variable attributes that can automatically grasp the characteristics of variables and then automatically select and analyze an appropriate statistical algorithm.

[0016] In view of the above, an object of the present invention is to provide an automated system for exploratory data analysis based on variable attributes, which provides a pre-screening process for a large number of variables and enables automatic application of an optimal data analysis algorithm based on such data characteristics (variable characteristics), thereby enabling data analysis to be performed well in accordance with the research objectives based on the data characteristics (variable characteristics) of the clinical researcher. [Means for solving the problem]

[0017] Concise and meaningful clinical research objectives are often expressed in concise and meaningful sentences (e.g., "Analysis of differences in cytokine changes between different treatment groups"). In order to carry out such research objectives, It is necessary to perform (a) normal distribution tests, (b) tests for identity of variances between groups, (c) post-hoc tests, and (d) decisions on nonparametric methods to use when the data do not follow normal distribution. This is a difficult process for clinical researchers, most of whom are not experts in statistics, to carry out directly.

[0018] Therefore, there is a great need for software that automatically selects the best-fit statistical algorithm to enable clinical researchers to best accomplish their research objectives based on their data characteristics (variable characteristics).

[0019] Therefore, the present invention provides an automated system for exploratory data analysis based on variable attributes, which is capable of automatically selecting the most suitable statistical algorithm for carrying out research objectives.

[0020] The variable attribute-based exploratory data analysis automation system of the present invention comprises: a variable attribute definition means for extracting characteristics of variables included in the data, classifying the variables according to the form and characteristics of the constituent values ​​of the variables, and defining attributes of the classified variables; an algorithm management means for storing and managing data analysis algorithms and providing the algorithms in response to a request from the data analysis control means; a data analysis control means for defining attributes of variables included in the data through the variable attribute definition means, selecting an algorithm for data analysis based on the variable attributes, and controlling execution of data analysis by the data analysis means; and a data analysis means for performing data analysis according to the algorithm set by the data analysis control means, and providing result information as a distribution analysis result table and a distribution analysis result graph, The variable attribute definition means classifies the data into continuous variables and categorical variables according to the form of the constituent values ​​of the variables included in the data, and classifies the categorical variables into a first classification variable including an ordinal variable and a nominal variable, and a second classification variable including a single measured variable, a repeatedly measured variable, and survival data according to the characteristics of the variables included in the data, and defines the attributes thereof; The data analysis control means comprises: a variable relation analysis control means for performing a variable relation analysis on the automatically classified variables according to a combination of their types by referring to variable attributes, and determining whether the analysis is a variable per se distribution analysis or a relation analysis between two or more variables; a variable per se distribution analysis control means for analyzing and controlling data for each of continuous variables and categorical variables when the analysis is a variable per se distribution analysis based on a result of the variable relation analysis control means; and a variable relation analysis control means for controlling data analysis by setting an algorithm by classifying when all variables are single-measured variables, when all variables are repeatedly-measured variables, when single-measured variables and repeatedly-measured variables are mixed, or when the data is survival data when the analysis is a relation analysis between two or more variables based on a result of the variable relation analysis control means.

[0021] The method for automating exploratory data analysis based on variable attributes of the present invention includes the steps of: Transformation for data analysis automation number A method for automating data analysis, comprising: a step of defining a variable attribute; and a step of selecting and setting an algorithm for data analysis based on the variable attribute, The step of defining one variable attribute includes: The method includes a process of classifying variables into continuous variables and categorical variables according to the form of the constituent values ​​of the variables included in the data, classifying categorical variables into first-class variables, which are classified as ordinal variables and nominal variables, and second-class variables, which include single measured variables, repeatedly measured variables, and survival data, according to the characteristics of the variables included in the data, and defining their attributes, The data analysis process involves selecting algorithms for data analysis based on variable attributes and performing data analysis. A variable relationship analysis process that performs a variable relationship analysis on the automatically classified variables according to the combination of their types and determines whether it is a distribution analysis of the variable itself or a relationship analysis between two or more variables; a variable distribution analysis step for analyzing the distribution of continuous variables and categorical variables based on the variable relationship analysis result of the variable relationship analysis step, and providing the results as a distribution analysis result table and a distribution analysis result chart; and a variable relationship analysis process for selecting and setting an algorithm according to whether the variables are all individually measured variables, all repeatedly measured variables, a mixture of individually measured variables and repeatedly measured variables, or survival data, in the case of a relationship analysis between two or more variables based on the result of the variable relationship analysis process, and providing a data analysis result. Effect of the Invention

[0022] According to the present invention, data analysis is possible by selecting an optimal algorithm based on variable attributes, and an automated data analysis system can be provided that is adaptive to variable data that is increasing in diversity as clinical processes and fields are subdivided in a variety of ways. [Brief description of the drawings]

[0023] [Figure 1] FIG. 1 is a block diagram showing a configuration of an automated system for exploratory data analysis based on variable attributes according to the present invention. [Diagram 2] 4 is a flowchart showing a data analysis control process in the present invention. [Diagram 3] 1 is a flowchart showing a continuous variable distribution data analysis control process according to the present invention. [Figure 4] 1 is a flowchart showing a categorical variable distribution data analysis control process in the present invention. [Figure 5a] FIG. 5a is a histogram showing the results of continuous variable data analysis in the present invention. [Figure 5b] FIG. 5b is a QQ plot showing the results of continuous variable data analysis in the present invention. [Figure 5c] FIG. 5c is a box plot showing the results of continuous variable data analysis in the present invention. [Figure 6] 1 is a bar plot showing the results of categorical variable distribution data analysis in the present invention. [Figure 7] 1 is a flowchart showing a process of analyzing the relationship between continuous variables and continuous variables in the present invention. [Figure 8] 2 is a flowchart illustrating the process of normal distribution testing for continuous variables in the present invention. [Figure 9]1 is a correlation scatter plot showing the results of a correlation analysis between continuous variables in the present invention. [Figure 10] 1 is a flowchart showing a data analysis control process in the case of a relationship between a continuous variable and a categorical variable in the present invention. [Figure 11] 1 is a flowchart showing a process of analyzing and controlling a relationship between categorical variables in the system of the present invention. [Figure 12] 1 is a flowchart showing the process of controlling the analysis of relational data between three or more individual measurement variables in the present invention. [Figure 13] 1 is a flow chart showing the data analysis control process when all variables are repeated measurements in the present invention. [Figure 14] 1 is a flowchart showing a data analysis control process when single measurement variables are mixed in the present invention. [Figure 15] 1 is a flowchart showing a process of controlling data analysis of survival samples in the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] The variable attribute-based exploratory data analysis automation system of the present invention comprises: The data analysis system includes a variable attribute definition means 10 that defines variable attributes for automating data analysis, an algorithm management means 20 in which data analysis algorithms are stored and managed, and that provides algorithms in response to requests from a data analysis control means 30, a data analysis control means 30 that defines attributes of variables included in data via the variable attribute definition means 10, selects an algorithm for data analysis based on the variable attributes, and controls the execution of data analysis by a data analysis means 40, and a data analysis means 40 that performs data analysis in accordance with the algorithm set by the data analysis control means 30, and provides information on the results.

[0025] The variable attribute definition means 10 includes a variable classification means 11 that extracts characteristics of variables contained in the data and classifies the variables according to the form and characteristics of the constituent values ​​of the variables, and an attribute definition means 12 that defines the attributes of the variables classified by the variable classification means 11.

[0026] The variable classification means 11 classifies the data into first classification variables including continuous variables and categorical variables according to the form of the constituent values ​​of the variables contained in the data, and second classification variables including single measured variables, repeatedly measured variables, and survival data according to the characteristics of the variables contained in the data.

[0027] The categorical variables are divided into ordinal variables and nominal variables.

[0028] The continuous variables are variables that indicate age, body mass index, height, etc., the ordinal variables are categorical variables that have an order, such as over 40 and under 40, and the nominal variables are categorical variables that do not have an order, such as male / female variables.

[0029] The single measurement variable means a variable that indicates a value measured once alone (e.g., health test values ​​measured in 2022), the repeatedly measured variable means a variable that indicates a value measured repeatedly within a certain period of time (e.g., hemoglobin values ​​measured before surgery / one month after surgery / one year after surgery), and the survival data means a variable that indicates sample status and period information (e.g., survival over 10 years of follow-up / censored / death information).

[0030] The data analysis control means 30 includes a variable relation analysis control means 31 which performs a variable relation analysis on the automatically classified variables according to a combination of their types by referring to the variable attributes, and judges whether the analysis is a variable per se distribution analysis or a relation analysis between two or more variables; a variable per se distribution analysis control means 32 which, when it is a variable per se distribution analysis based on the variable relation analysis result of the variable relation analysis control means 31, analyzes and controls data for each of continuous variables and categorical variables; and, when it is a relation analysis between two or more variables based on the variable relation analysis result of the variable relation analysis control means 31, analyzes and controls data by classifying the data into when all variables are single-measured variables, when all variables are repeatedly-measured variables, when single-measured variables and repeatedly-measured variables are mixed, and when it is survival data, by setting an algorithm.

[0031] The data analysis means 40 includes a variable distribution analysis means 41 for analyzing data for each of continuous variables and categorical variables under the control of the variable distribution analysis control means 32 and providing the results as a distribution analysis result table and a distribution analysis result graph, and an inter-variable relationship analysis means 42 for analyzing data by executing a set algorithm for when all variables are single-measurement variables, when all variables are repeatedly-measured variables, when single-measurement variables and repeatedly-measured variables are mixed, or when data is survival data, depending on the inter-variable relationship under the control of the inter-variable relationship analysis control means 33, and providing the results.

[0032] The data analysis control means 30 includes a normal distribution test means 34 for continuous variables that calculates a significance probability value (p-value, p1) by performing a Lilliefors test for the continuous variables, calculates a significance probability value (p-value, p2) by performing a Shapiro-Wilk test, and calculates a normal distribution test result for the continuous variables by comparing p1 and p2 with a reference value α.

[0033] The normal distribution test result for the continuous variable is determined not to follow a normal distribution if the condition p1<α (AND or OR) p2<α is satisfied, and determined to follow a normal distribution if the condition is not satisfied.

[0034] The normal distribution test means 34 for the continuous variable may include a reference value setting means 34a that provides a reference value setting process so that a user can set the reference value α.

[0035] The variable distribution analysis means 41 can provide continuous variable distributions showing the results of continuous variable data analysis as a distribution analysis result table, a histogram, a QQ plot, and a box plot, and can provide categorical variable distributions showing the results of data analysis as a distribution analysis result table and a bar graph.

[0036] The values ​​calculated in the continuous variable distribution analysis result table are: (a) Total N: The total number of samples included in the data (b) Valid N(%): The number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Min~Max: Minimum to maximum value of the variable (e) Mean ± standard deviation: The mean ± standard deviation of the variable (f) Mean(95%CIs): Mean of the variable (95% confidence interval) (g)Median(IQR): The median (inter-quartile range) of the variable (h)Skewness: Skewness of the variable (i) Kurtosis: The kurtosis of the variable (j) Lilliefors test for normality, p value: The significance probability value (p value) of the normal distribution of the variable in question as a result of testing it with the Lilliefors method (k) Shapiro-Wilk test for normality, p value: p value of the normal distribution of the variable tested using the Shapiro-Wilk method The values ​​calculated in the categorical variable distribution analysis result table are: (a) Total N: The total number of samples included in the data (b) Valid N(%): The number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Subgroup: Name of the subgroup included in the categorical variable (e) N(%): The number and percentage of samples included in the subgroup (f) 95%CI: 95% confidence interval for the percentage of samples included in the subgroup The variable relationship analysis control means 33 is This is a means of providing a process for analyzing and controlling data by setting algorithms according to whether the data are all individually measured variables, all repeatedly measured variables, a mixture of individually and repeatedly measured variables, or survival data.

[0037] [When all variables are individually measured] When all variables are single measurement variables, data analysis is performed by dividing the data into relationship analysis between two variables and relationship analysis between three or more variables. When the data is a relationship analysis between two variables, data analysis is performed according to the relationship between continuous variables and continuous variables, the relationship between continuous variables and categorical variables, and the relationship between categorical variables and categorical variables. When the data is a relationship analysis between three or more variables, data analysis is performed by dividing the data into when all variables are continuous, when all variables are categorical, and when continuous and categorical variables are mixed.

[0038] [Relationship between continuous variables] In the case of the relationship between continuous variables, the data analysis control process includes the steps of: performing a normal distribution test on the two continuous variables; and judging whether both variables follow a normal distribution based on the result of the normal distribution test. If they do follow a normal distribution, the data analysis control is performed by setting a Pearson correlation analysis; if they do not follow a normal distribution, the data analysis control is performed by setting a Spearman correlation analysis; and providing the analysis results as an analysis result table and an analysis result graph.

[0039] The numerical values ​​presented in the analysis result table provided as the analysis result of the relationship between the continuous variables are: (a) Correlation coefficient (b) 95% confidence interval of the correlation coefficient (c) The significance probability value (p-value) calculated as a result of testing whether the correlation coefficient is 0 The analysis result diagram provided as the analysis result of the relationship between the continuous variables can be composed of a correlation scatter plot in which x-axis and y-axis variables are set and a regression curve is displayed.

[0040] [Relationship between continuous and categorical variables] In the case of a relationship between a continuous variable and a categorical variable, the data analysis control process Data analysis is performed by dividing the data into mean difference analysis, correlation analysis, and dichotomous response prediction performance analysis. In the case of the mean variance analysis, the steps are: extracting the number of categorical variable subgroups (m); and performing a normal distribution test on the continuous variables when the number of categorical variable subgroups (m) is two and when the number of categorical variable subgroups is three or more; judging whether the continuous variables follow a normal distribution when the number of categorical variable subgroups (m) is two; setting a Wilcoxon rank-sum test when the continuous variables do not follow a normal distribution when the number of categorical variable subgroups (m) is two; testing whether the variances of the subgroups are the same by performing a Levene's test when the number of categorical variable subgroups (m) is two and the continuous variables follow a normal distribution when the number of categorical variable subgroups (m) is two; setting a "Student's Test" if the variances of the subgroups are the same; and setting a "Welch's T test" if the variances of the subgroups are not the same to control the data analysis; and performing a one-way analysis of variance (1-way ANOVA) when the number of categorical variable subgroups (m) is three or more and the continuous variables follow a normal distribution. and setting up a Kruskal-Wallis H test to control the data analysis if the continuous variables do not follow a normal distribution.

[0041] And, in the case of the relationship between the continuous variable and the categorical variable, the data analysis control process The method further includes a process of performing a post-hoc analysis after the data analysis using the 1-way ANOVA and Kruskal-Wallis H test, The algorithms used in the post-hoc validation performed after the data analysis by the 1-way ANOVA include the Bonferroni test, the Tukey test, the Scheffe test, and the Dunnett test, and the algorithms used in the post-hoc validation performed after the data analysis by the Kruskal-Wallis H test include the Bonferroni test, the FDR (False Discovery rate), and the Dunn's test.

[0042] The dichotomous response prediction performance analysis includes a process of extracting the number of categorical variable subgroups (m) and controlling the data analysis by setting up an ROC curve analysis when the number of extracted categorical variable subgroups (m) is two.

[0043] The correlation analysis includes the steps of: performing a normal distribution analysis on a continuous variable; determining whether the continuous variable follows a normal distribution; determining whether the categorical variable is ordinal if the continuous variable follows a normal distribution; setting a Polyserial correlation analysis to control data analysis if the categorical variable is ordinal, and setting a Point polyserial correlation analysis to control data analysis if the categorical variable is not ordinal; determining whether the categorical variable is ordinal if the continuous variable does not follow a normal distribution; setting a Polychoric correlation analysis to control data analysis if the categorical variable is ordinal, and setting a Rank polyserial correlation analysis to control data analysis if the categorical variable is not ordinal.

[0044] [Relationship between categorical variables and categorical variables] In the case where the relationship is between categorical variables, the data analysis control process The method includes a step of controlling data analysis by dividing the number of subgroups (m, n) included in the categorical variables into two cases and other cases, and dividing the number of subgroups (m, n) included in the categorical variables into two cases, namely, an independence test, and a trend test, and a correlation analysis, when the number of subgroups (m, n) included in the categorical variables is not two; The independence test includes the steps of preparing a 2×2 cross table combining two subgroups of a categorical variable, preparing an expectation table under the assumption that the categorical variables are independent, and judging whether the number of cells with an expectation value < 5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells. If the number of cells with an expectation value < 5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells, taking into consideration the computational capacity of the computer, Fisher's exact test is set if the computational capacity of the computer is sufficient, and if the computational capacity of the computer is not sufficient, Chi-squared test with Yates correction is set. If the number of cells with an expectation value < 5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is less than 25% of the total number of cells, judging whether the cross table is in the form of 2×2, and if the cross table is in the form of 2×2, Chi-squared test with Yates correction is set, and if the cross table is not in the form of 2×2, Chi-squared test with Yates correction is set. and setting up the test and controlling the data analysis.

[0045] The trend test includes the steps of creating an m×n cross table combining two subgroups of categorical variables and determining whether the conditions m≧3 and n≧3 are met; and selecting and setting a Linear by Linear Assocation Test if the conditions m≧3 and n≧3 are met, and setting a Cochran's Q test to control the data analysis if the conditions m≧3 and n≧3 are not met.

[0046] The correlation analysis includes a process of determining whether all categorical variables are ordinal, and if so, selecting and setting Polychoric correlation analysis; if not all categorical variables are ordinal, determining whether all categorical variables are nominal, and if so, selecting and setting Cramer's V analysis; if not all categorical variables are nominal, setting Rank polyserial correlation analysis to control data analysis.

[0047] [Analysis of relationships between three or more all independently measured variables] The relationship analysis data analysis control process among all three or more independent measurement variables includes: When three or more variable groups are all continuous, when they are all categorical, or when continuous and categorical variables are mixed, data analysis is performed by dividing them into categories. If all data are continuous, the process involves setting up PCA (Principle Component Analysis) and controlling the data analysis. When continuous variables and categorical variables are mixed, the method includes the steps of: using each of the continuous variables as a dependent variable and setting up a univariable linear regression to analyze the individual influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and setting up a multivariable linear regression to analyze the integrated influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and setting up an ANCOVA (Analysis of Covariance) to control the data analysis when there is a categorical variable among the remaining variables; and using each of the continuous variables as a dependent variable and setting up a 2-way ANOVA to control the data analysis when all of the remaining variables are categorical variables.

[0048] When all the variables are categorical, the process includes determining whether there are dichotomous variables; if there are dichotomous variables, using each of the dichotomous variables as a dependent variable, setting up univariate binary logistic regression to analyze the individual influence of the remaining variables on the dichotomous variables, setting up multivariable binary logistic regression to analyze the integrated influence of the remaining variables on the dichotomous variables; constructing a dichotomous response prediction model; and controlling the validation analysis of the constructed dichotomous response prediction model. When there are no dichotomous variables and all variables are ternary or higher, each of the variables is used as a dependent variable, the ternary variables are assumed to be ordinal variables, and the ternary variables are assumed to be nominal variables. For each of the ternary variables, an analysis of the individual influence and the integrated influence of the remaining variables on the ternary ordinal variable and the ternary nominal variable is performed. This includes a process of setting up a univariable ordinal logistic regression to analyze the individual influence of the remaining variables on the ternary ordinal variable, a process of setting up a multicariable ordinal logistic regression to analyze the integrated influence of the remaining variables on the ternary ordinal variable, a process of setting up a univariable nominal logistic regression to analyze the individual influence of the remaining variables on the ternary nominal variable, and a process of setting up a multicariable nominal logistic regression to analyze the integrated influence of the remaining variables on the ternary nominal variable.

[0049] The validation analysis of the dichotomous reaction prediction model is performed by a discrimination aspect prediction performance analysis, a calibration aspect prediction performance analysis, and a model performance cross-validation analysis.

[0050] The indicators used in the discrimination aspect prediction performance analysis are: (a) Performance analysis index (including 95% confidence interval) ·UC(95%CI) Sensitivity, Specificity ·PPV(positive predictive value),NPV(negative predictive value) ·ACC(accuracy), MIS(miss-classification rate) ·FPR(False Positive Rate), FNR(False Negative Rate), FDR(False Discovery Rate), FOR(False Omission Rate) ·LR+(Positive Likelihood Ratio), LR-(Negative Likelihood Ratio), DOR(Diagnostic Odds Ratio) (b) Visualization of performance analysis results ROC Curve It is.

[0051] The indices used in the Calibration aspect performance analysis are: (a) Performance analysis index ·AIC(Akaike Information Criterion) ·BIC(Bayes Information Criterion) ·Nagelkerke R2 ·Hosmer-Lemeshow test P value Brier score ·Spigelhalter Z score with P value ·Linear regression line in calibration plot Intercept, 95% confidence interval and p-value Slope, 95% confidence interval and p-value (b) Visualization of performance analysis results Calibration plot Decile plot Calibration belt It is.

[0052] The model performance cross-validation analysis includes: (a) Methodology used for performance cross-validation ·LOOCV(leave-one-out cross-validation) K-fold CV (cross validation) Permutation test Bootstrapping (b) Visualization of cross-validation results ·ROC curve Calibration plot Decile plot Calibration belt It is.

[0053] [Analysis when all variables are repeated measurements] The whole repeated measurement variable analysis data analysis control process includes: Data analysis is performed by dividing the cases into those with two repeated measurement variables and those with three or more repeated measurement variables, and those with continuous measurement variables and those with categorical measurement variables. When there are two repeated measurement variables and the repeated measurement variables are continuous variables, the method includes the steps of: performing normal distribution analysis on the continuous variables to determine whether the continuous variables follow a normal distribution; if they do follow a normal distribution, setting up a paired sample T test; if they do not follow a normal distribution, setting up a Wilcoxon signed-rank test to control the data analysis.

[0054] In addition, when there are two repeatedly measured variables and the repeatedly measured variables are continuous variables, the method may further include the step of controlling data analysis by setting an Intraclass Correlation Coefficient (ICC) analysis.

[0055] When there are two repeated measurement variables and the repeated measurement variables are categorical variables, the method includes the steps of: when the number of subgroups (m, n) included in the categorical variables is two (m=2, n=2), setting McNemar's test and Cohens' Kappa to control the data analysis; when the number of subgroups (m, n) included in the categorical variables is two or more (m=2, n≧3), setting Cochran-Armitage test for trend to control the data analysis; and when the number of subgroups (m, n) included in the categorical variables is three or more (m≧3, n≧3), setting McNemar-Bowker test to control the data analysis.

[0056] When there are three or more repeated measurement variables and the repeated measurement variables are continuous variables, the process includes setting up Linear mixed effect model analysis, GEE (Generalized Estimating Equation) analysis, and Repeated measures 1-way ANOVA to control the data analysis.

[0057] When there are three or more repeatedly measured variables and the repeatedly measured variables are categorical variables, the process includes setting up a Generalized mixed effect model analysis, or GEE (Generalized Estimating Equation) analysis to control the data analysis.

[0058] [Analysis when single measured variables are mixed] The data analysis control process when the single measurement variables are mixed is as follows: The repeated measurement variables are classified into two types: continuous variables and categorical variables. When the repeated measurement variables are continuous variables, a Linear mixed effect model analysis or a GEE (Generalized Estimating Equation) analysis is set up to control the data analysis, and when the repeated measurement variables are continuous variables and the single measurement variable is a categorical variable, a Repeated measures 2-way ANOVA is set up to control the data analysis.

[0059] When the repeated measurement variables are categorical variables, the process includes setting up a generalized mixed effect model analysis, or GEE (Generalized Estimating Equation) analysis, to control the data analysis.

[0060] [Analysis of Survival Materials] The survival data analysis control process includes: The cases are divided into those with only survival time and event occurrence variables and those with survival time, event occurrence variables, and single measurement data. When only survival time and event occurrence variables are available, Kaplan-Meier curve analysis is set up to control data analysis; when survival time, event occurrence data and single measurement data are available, it includes the steps of setting up univariable cox proportional hazards regression to control the analysis of the individual influence of single measurement variables on survival, setting up multivariable cox proportional hazards regression to analyze and control the integrated influence of single measurement variables on survival, and setting up Kaplan-Meier curve analysis to control Log rank test (comparison of survival probability differences between detailed categories).

[0061] And, the survival data analysis control process is The method may further include a process of constructing a survival probability prediction model, and a process of performing a discrimination aspect prediction performance analysis at time=t, a calibration aspect prediction performance analysis at time=t, and a survival probability prediction model cross-validation analysis to analyze and control the survival probability prediction model.

[0062] The indicators used in the discrimination aspect prediction performance analysis at Time=t are, as described above, (a) Performance analysis index (including 95% confidence interval) Area under the curve (95%CI) Sensitivity, Specificity ·PPV(positive predictive value), NPV(negative predictive value) ·ACC(accuracy), MIS(miss-classification rate) ·FPR(False Positive Rate), FNR(False Negative Rate), FDR(False Discovery Rate), FOR(False Omission Rate) ·LR+(Positive Likelihood Ratio), LR-(Negative Likelihood Ratio), DOR(Diagnostic Odds Ratio) (b) Visualization of performance analysis results ·ROC curve It is.

[0063] The indexes used in the calibration performance analysis at Time=t are: (a) Performance analysis index ·AIC(Akaike Information Criterion) ·BIC(Bayes Information Criterion) ·Nagelkerke R2 ·Hosmer-Lemeshow test P value Brier score ·Spigelhalter Z score with P value ·Linear regression line in calibration plot Intercept, 95% confidence interval and p-value Slope, 95% confidence interval and p-value (b) Visualization of performance analysis results Calibration plot Decile plot Calibration belt It is.

[0064] The survival probability prediction model performance cross-validation at time=t is (a) Methodology used for performance cross-validation ·LOOCV(leave-one-out cross-validation) K-fold CV (cross validation) Permutation test Bootstrapping (b) Visualization of cross-validation results ·ROC curve Calibration plot Decile plot Calibration belt It is.

[0065] Next, a data analysis automation operation process of the exploratory data analysis automation system based on variable attributes according to the present invention will be described.

[0066] To automate data analysis, the variable attribute definition means 10 defines the attributes of the variables contained in the data.

[0067] The data analysis control means 30 classifies the variables contained in the data based on the variable attributes via the variable attribute definition means 10 and controls the data analysis accordingly, and also selects and sets the algorithm to be used for data analysis by the data analysis means 40 from the algorithm management means 20 based on the variable attributes, thereby controlling the data analysis.

[0068] The variable classification means 11 of the variable attribute definition means 10 extracts the characteristics of the variables contained in the data, and classifies the variables according to the form and characteristics of the constituent values ​​of the variables.

[0069] The variable classification means 11 classifies variables into first classification variables including continuous variables and categorical variables according to the form of the constituent values ​​of the variables contained in the data, and second classification variables including single measurement variables, repeated measurement variables, and survival data according to the characteristics of the variables contained in the data.

[0070] In this case, the categorical variables can be divided into ordinal variables and nominal variables. The continuous variables are variables indicating age, body mass index, height, etc., the ordinal variables are categorical variables having an order such as over 40 years old and under 40 years old, and the nominal variables are categorical variables having no order such as male / female variables.

[0071] The single measurement variable means a variable that indicates a value measured once alone (e.g., health test values ​​measured in 2022), the repeatedly measured variable means a variable that indicates a value measured repeatedly within a certain period of time (e.g., hemoglobin values ​​measured before surgery / one month after surgery / one year after surgery), and the survival data means a variable that indicates the status and period information of the sample (e.g., survival over 10 years of follow-up / censored / death information).

[0072] The data analysis control means 30 performs a variable relationship analysis on the automatically classified variables according to the combination of their types by referring to the variable attributes, and analyzes the data based on the result of the variable relationship analysis. Here, it can be determined whether the analysis is a distribution analysis of the variables themselves or an analysis of the relationship between two or more variables.

[0073] FIG. 2 is a flowchart showing a process of analyzing data based on variable attributes for variable data in the system of the present invention.

[0074] In the case of variable data, the distribution analysis of variables is performed for (A) continuous variable distribution, or (B) categorical variable distribution. In the case of relationship analysis between two or more variables, data analysis is performed according to the relationship between the variables.

[0075] First, in the case of a distribution analysis of the variables themselves, as shown in Figures 3 and 4, data is analyzed for each of the continuous variables and categorical variables, and the results are provided as a distribution analysis result table and a distribution analysis result chart.

[0076] The variable distribution analysis means 41 can provide continuous variable distributions showing the results of continuous variable data analysis as a distribution analysis result table, and histograms, QQ plots, and box plots as shown in Figures 5a, 5b, and 5c, and can provide categorical variable distributions showing the results of data analysis as a distribution analysis result table and a bar graph as shown in Figure 6.

[0077] On the other hand, when analyzing the relationship between two or more variables, the data is analyzed by setting algorithms according to whether they are all individually measured variables, all repeatedly measured variables, a mixture of individually and repeatedly measured variables, or survival data.

[0078] [When all variables are individually measured] When all variables are measured individually, data analysis is performed by dividing the data into analysis of the relationship between two variables and analysis of the relationship between three or more variables.

[0079] In the case of relationship analysis between two variables, data analysis is performed according to (C) the relationship between continuous variables and continuous variables, (D) the relationship between continuous variables and categorical variables, and (E) the relationship between categorical variables and categorical variables. In the case of relationship analysis between three or more variables, (F) the data analysis is performed by classifying when all variables are continuous variables, when all variables are categorical variables, and when there is one or more continuous variables and one or more categorical variables.

[0080] [Relationship between continuous variables] FIG. 7 is a flowchart showing the process of analyzing the relationship between continuous variables in the system of the present invention.

[0081] In the case of a relationship between two continuous variables, a normal distribution test for the two continuous variables is performed.

[0082] FIG. 8 is a flow chart illustrating the normal distribution testing process for continuous variables.

[0083] For continuous variables, the significance probability value (p-value, p1) is calculated by performing the Lilliefors test, and the significance probability value (p-value, p2) is calculated by performing the Shapiro-Wilk test.

[0084] The significance probability values ​​p1 and p2 obtained as described above are compared with a reference value α to calculate the normal distribution test result for the continuous variable.

[0085] The normal distribution test result for the continuous variable is determined not to follow a normal distribution if the condition p1<α (AND or OR) p2<α is satisfied, and determined to follow a normal distribution if the condition is not satisfied.

[0086] At this time, the reference value α can be set to a value such as 0.05 or 0.01 by the user via the reference value setting means 34a.

[0087] After conducting the continuous variable normal distribution test as described above, it is judged whether all the data follow normal distribution. If they do, Pearson correlation analysis is set up to control the data analysis. If they do not follow normal distribution, Spearman correlation analysis is set up to control the data analysis. The analysis results are provided as an analysis result table and analysis result chart.

[0088] The numerical values ​​presented in the analysis result table provided as the analysis result of the relationship between the continuous variables are: (a) Correlation coefficient (b) 95% confidence interval of the correlation coefficient (c) It is equal to the significance value (p-value) calculated as a result of testing whether the correlation coefficient is 0.

[0089] The analysis result diagram provided by the analysis result of the relationship between the continuous variables may be a correlation scatter plot in which x-axis and y-axis variables are set and a regression curve is displayed, as shown in FIG. 9.

[0090] [Relationship between continuous and categorical variables] In the case of (D) the relationship between continuous variables and categorical variables, data analysis is performed by classifying into mean difference analysis, correlation analysis, and dichotomous response prediction performance analysis.

[0091] FIG. 10 is a flowchart showing the data analysis control process in the case of a relationship between continuous variables and categorical variables.

[0092] In the mean variance analysis, the number of categorical variable subgroups (m) is extracted, and when the number of categorical variable subgroups (m) extracted is two, or when the number of categorical variable subgroups is three or more, a normal distribution test as shown in FIG. 8 is performed on the continuous variables, and an algorithm is set based on the results.

[0093] When the number of categorical variable subgroups (m) is two, determine whether the continuous variable follows a normal distribution. If the continuous variable does not follow a normal distribution, set the Wilcoxon rank-sum test. If the continuous variable follows a normal distribution, perform Levene's test to test whether the variances of each subgroup are the same. If the variances of each subgroup are the same, set the "Student's Test." If the variances of each subgroup are not the same, set the "Welch's T test" to perform data analysis.

[0094] Then, when the number of categorical variable subgroups (m) is three or more, if the continuous variables follow a normal distribution, one-way ANOVA is set up to control the data analysis; if the continuous variables do not follow a normal distribution, the Kruskal-Wallis H test is set up to perform the data analysis.

[0095] Next, we performed a post-hoc analysis after the data analysis using the 1-way ANOVA and Kruskal-Wallis H test. Post-hoc validation after data analysis by 1-way ANOVA is performed by setting up Bonferroni test, Tukey test, Scheffe test, and Dunnett test, and post-hoc validation after data analysis by the Kruskal-Wallis H test is performed by setting up Bonferroni test, FDR (False Discovery rate), and Dunn's test.

[0096] The correlation analysis involves performing a normal distribution analysis on the continuous variables as shown in FIG. 8 to determine whether the continuous variables follow a normal distribution.

[0097] When the continuous variables follow a normal distribution, determine whether the categorical variables are ordinal; if the categorical variables are ordinal, set Polyserial correlation analysis to control the data analysis; if the categorical variables are not ordinal, set Point polyserial correlation analysis to perform the data analysis.

[0098] If the continuous variable does not follow a normal distribution, determine whether the categorical variable is ordinal. If the categorical variable is ordinal, set up Polychoric correlation analysis to analyze the data. If the categorical variable is not nominal, set up Rank polyserial correlation analysis to analyze the data.

[0099] Then, in the dichotomous response prediction performance analysis, the number of categorical variable subgroups (m) is extracted, and when the number of categorical variable subgroups (m) is two, ROC curve analysis is set up to perform data analysis.

[0100] [Relationship between categorical variables and categorical variables] FIG. 11 is a flowchart showing a process of controlling analysis of the relationship between categorical variables in the system of the present invention.

[0101] In the case of the relationship between categorical variables (E) and categorical variables, The data is analyzed by dividing the data into cases where the number of subgroups (m, n) included in the categorical variables is always two and cases where it is not, and by dividing the data into cases where the number of subgroups (m, n) included in the categorical variables is always two, then an independence test is performed, and cases where the number of subgroups (m, n) included in the categorical variables is not two, then an independence test, trend test, and correlation analysis are performed.

[0102] The independence test involves creating a 2 x 2 cross table combining two subgroups of a categorical variable, creating an expectation table under the assumption that the categorical variables are independent, and determining whether the number of cells with an expectation value < 5 in the m x n cross table for the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells.

[0103] When the number of cells with an expected value < 5 in an m x n crosstabulation of the number of subgroups (m, n) included in a categorical variable is 25% or more of the total number of cells, data analysis is performed taking into consideration the computing capacity of the computer. If the computing capacity of the computer is sufficient, Fisher's exact test is set, and if the computing capacity of the computer is not sufficient, Chi-squared test with Yates correction is set.

[0104] If the number of cells with an expected value < 5 in an m x n cross table of the number of subgroups (m, n) included in a categorical variable is less than 25% of the total number of cells, determine whether the cross table is in a 2 x 2 format. If the cross table is in a 2 x 2 format, set up a Chi-squared test with Yates correction. If the cross table is not in a 2 x 2 format, set up a Chi-squared test and perform data analysis.

[0105] This is due to statistical analysis which shows that if the expected value is 25% or more, the data can be considered to have a chi-squared distribution, but if the expected value is less than 25%, the data does not follow a chi-squared distribution.

[0106] For example, in the case of the way of eating Tangsuyuk (Korean sweet and sour pork), which is composed of a gender variable and two categorical variables of the way of eating Tangsuyuk, Male: dipping meat in sauce 25, pouring sauce over meat 25, Female: dipping meat in sauce 25, pouring sauce over meat 25, it is analyzed that the way of eating Tangsuyuk does not differ depending on gender, and this is statistically analyzed as "gender and way of eating Tangsuyuk are independent of each other."

[0107] However, if the observed patterns are as follows: Men: dipping meat in sauce (10), pouring sauce over meat (40), and Women: dipping meat in sauce (35), pouring sauce over meat (15), it can be concluded that men prefer pouring sauce over meat and women prefer dipping meat in sauce.

[0108] Therefore, since there is a relationship between gender and the way of eating tangsuyuk, it can be said that they are "not independent."

[0109] The trend test involves creating an m×n cross table combining two subgroups of categorical variables, determining whether the conditions m≧3 and n≧3 are met, and if these conditions are met, selecting and setting the Linear by linear assciation test, and if these conditions are not met, setting the Cochran's Q test to perform data analysis.

[0110] Here, since 2x2 only has a dichotomous form such as (0, 1) or (Yes, No), and "tendency" and "independence" have essentially the same meaning, only independence is analyzed. However, when both m and n are 3 or more, both the independence test and trend test as described above can be performed.

[0111] The trend test can be explained using the examples of age group variables and OTP usage pattern variables (not using, using only one, using two or more). In other words, it can be considered to conduct a "trend" analysis such as increase or decrease, as if the number of OTT service users increases, decreases, or there is no difference as the age group increases.

[0112] The correlation analysis is carried out by determining whether all categorical variables are ordinal, and if so, selecting and setting Polychoric correlation analysis; if not all categorical variables are ordinal, determining whether all categorical variables are nominal, and if so, selecting and setting Cramer's V analysis; if not all categorical variables are nominal, setting Rank polyserial correlation analysis to perform data analysis.

[0113] [Analysis of relationships between three or more all independently measured variables] The (F) analysis of the relationship between all three or more independently measured variables includes: When three or more variable groups are all continuous, when they are all categorical, or when there is a mixture of continuous and categorical variables, data analysis is performed by dividing them into groups.

[0114] FIG. 12 is a flow chart showing the process of controlling the analysis of relational data among three or more total independently measured variables in the present invention.

[0115] If all the data are continuous, use PCA (Principa Component Analysis) to analyze the data. When continuous and categorical variables are mixed, each of the continuous variables is used as the dependent variable and univariable linear regression is set up to analyze the individual influence of the remaining variables on the continuous variables; each of the continuous variables is used as the dependent variable and multivariable linear regression is set up to analyze the integrated influence of the remaining variables on the continuous variables; when each of the continuous variables is used as the dependent variable and there is a categorical variable among the remaining variables, ANCOVA (Analysis of Covariance) is set up to perform data analysis; when each of the continuous variables is used as the dependent variable and all of the remaining variables are categorical variables, 2-way ANOVA is performed to perform data analysis.

[0116] When all variables are categorical, data analysis is performed according to the presence or absence of dichotomous variables. When dichotomous variables are present, each of the dichotomous variables is used as the dependent variable, and the individual influence of the remaining variables on the dichotomous variables is analyzed, and the integrated influence of the remaining variables on the dichotomous variables is analyzed.

[0117] Univariable binary logistic regression is set up to analyze the individual influence of the remaining variables on the dichotomous variables, and multivariable binary logistic regression is set up to analyze the joint influence of the remaining variables on the dichotomous variables.

[0118] In addition, a dichotomous reaction prediction model can be constructed, and a verification analysis can be performed on the constructed dichotomous reaction prediction model.

[0119] The validation analysis of the dichotomous reaction prediction model is performed by a discrimination aspect prediction performance analysis, a calibration aspect prediction performance analysis, and a model performance cross-validation analysis.

[0120] [Analysis when all variables are repeated measurements] The (G) all repeated measures variable analysis is Data analysis is performed by dividing the data into cases where there are two repeatedly measured variables and cases where there are three or more repeatedly measured variables, and into cases where the repeatedly measured variables are continuous variables and cases where the repeatedly measured variables are categorical variables.

[0121] FIG. 13 is a flow chart showing the data analysis control process when all variables are repeated measurements in the system of the present invention.

[0122] When there are two repeated measurement variables and the repeated measurement variables are continuous variables, a normal distribution analysis is performed on the continuous variables to determine whether the continuous variables follow a normal distribution. If they do follow a normal distribution, a paired sample T test is set up, and if they do not follow a normal distribution, a Wilcoxon signed-rank test is set up to perform data analysis.

[0123] In addition, when there are two repeated measurement variables and the repeated measurement variables are continuous variables, an ICC (Intraclass Correlation Coefficient) analysis can be set to perform data analysis.

[0124] When there are two repeated measurement variables and the repeated measurement variables are categorical variables, if the number of subgroups (m, n) contained in the categorical variables is all two (m=2, n=2), McNemar's test and Cohens' Kappa are used for data analysis; if the number of subgroups (m, n) contained in the categorical variables is two or three or more (m=2, n≧3), Cochran-Armitage test for trend is used for data analysis; and if the number of subgroups (m, n) contained in the categorical variables is all three or more (m≧3, n≧3), McNemar-Bowker test is used for data analysis.

[0125] When there are three or more repeated measurement variables and the repeated measurement variables are continuous, data analysis is performed using Linear mixed effect model analysis, GEE (Generalized Estimating Equation) analysis, and Repeated measures 1-way ANOVA.

[0126] When there are three or more repeated measurement variables and the repeated measurement variables are categorical variables, generalized mixed effect model analysis and GEE (Generalized Estimating Equation) analysis are set up to perform data analysis.

[0127] [Analysis when single measured variables are mixed] FIG. 14 is a flow chart showing the data analysis control process in the system of the present invention when single measurement variables are mixed.

[0128] The analysis of (H) above when single measurement variables are mixed is divided into cases where the repeatedly measured variables are continuous variables and cases where the repeatedly measured variables are categorical variables, When the repeated measurement variables are continuous variables, linear mixed effect model analysis and GEE (Generalized Estimating Equation) analysis are used to analyze the data. When the repeated measurement variables are continuous variables and the single measurement variable is a categorical variable, repeated measures 2-way ANOVA is used to analyze the data.

[0129] When the repeated measurement variables are categorical variables, data analysis is performed using Generalized mixed effect model analysis and Generalized Estimating Equation (GEE) analysis.

[0130] [Analysis of Survival Materials] FIG. 15 is a flow chart showing the process of controlling data analysis of survival samples in the system of the present invention.

[0131] The (I) survival data analysis is divided into a case where only survival time and event occurrence variables are present and a case where survival time, event occurrence variables and single measurement data are present; When only survival time and event occurrence variables are available, Kaplan-Meier curve analysis is set up to control data analysis; when survival time, event occurrence data and single measurement data are available, univariable cox proportional hazards regression is set up to analyze the individual influence of single measurement variables on survival, and multivariable cox proportional hazards regression is set up to analyze the integrated influence of single measurement variables on survival.

[0132] In addition, Kaplan-Meier curve analysis and Log rank test (comparison of survival probability differences between detailed categories) will be performed.

[0133] And, the survival data analysis control process is A survival probability prediction model is constructed, and the process of analyzing the survival probability prediction model is carried out by performing a discrimination aspect prediction performance analysis at time=t, a calibration aspect prediction performance analysis at time=t, and a survival probability prediction model cross-validation analysis.

[0134] In this way, according to the system of the present invention, the number of cases required for data analysis is classified based on variable attributes, the algorithm required for data analysis is selected and set in accordance with this classification, and data analysis is automatically performed using this, thereby providing an automated system that performs data analysis that is optimal for the user's desired purpose.

[0135] On the other hand, the method for automating exploratory data analysis based on variable attributes of the present invention using such a system is as follows: Transformation for automated statistical analysis number A variable attribute definition process for defining attributes; and a data analysis process for selecting and setting an algorithm for data analysis based on the variable attributes, and performing data analysis; The variable attribute definition step of defining variable attributes for automating statistical analysis includes: This involves the process of classifying variables into first classifications, including continuous variables and categorical variables, according to the form of the constituent values ​​of the variables contained in the data, and second classifications, including single measured variables, repeatedly measured variables, and survival data, according to the characteristics of the variables contained in the data, and defining their attributes.

[0136] The categorical variables are divided into ordinal variables and nominal variables.

[0137] The continuous variables are variables that indicate age, body mass index, height, etc., the ordinal variables are categorical variables that have an order, such as over 40 and under 40, and the named variables are categorical variables that do not have an order, such as male / female variables.

[0138] The single measurement variable means a variable indicating a value measured once alone (e.g., health test values ​​measured in 2022), the repeatedly measured variable means a variable indicating a value measured repeatedly within a certain period of time (e.g., hemoglobin values ​​measured before surgery / one month after surgery / one year after surgery), and the survival data means a variable indicating sample status and period information (e.g., survival / censored / death information for 10 years of follow-up period).

[0139] The data analysis step of selecting an algorithm for data analysis based on the variable attributes and performing data analysis includes: As shown in Figure 2, a variable relationship analysis process is performed on the automatically classified variables according to their type combinations to determine whether the analysis is a distribution analysis of the variable itself or a relationship analysis between two or more variables. a variable distribution analysis step for analyzing the distribution of continuous variables and categorical variables based on the variable relationship analysis result of the variable relationship analysis step, and providing the results as a distribution analysis result table and a distribution analysis result chart; and a variable relationship analysis process for selecting and setting an algorithm according to whether the variables are all individually measured variables, all repeatedly measured variables, a mixture of individually measured variables and repeatedly measured variables, or survival variables, based on the result of the variable relationship analysis process, in the case of a relationship analysis between two or more variables, and providing a data analysis result.

[0140] The process of analyzing the distribution of the variables themselves includes a process of testing normal distribution of continuous variables, which includes a process of calculating a significant probability value (p-value: p1) by performing a Lilliefors test, a process of calculating a significant probability value (p-value: p2) by performing a Shapiro-Wilks test, and a process of calculating a normal distribution test result for the continuous variables by comparing a set reference value α with p1 and p2, as shown in FIG. 8.

[0141] The normal distribution test result for the continuous variable is determined not to follow a normal distribution if the condition p1<α (AND or OR) p2<α is satisfied, and determined to follow a normal distribution if the condition is not satisfied.

[0142] The set reference value α may be a value set by a user.

[0143] In the process of analyzing the distribution of the variables themselves, for continuous variable distribution, data analysis results can be provided as a distribution analysis result table and histograms, QQ plots, and box plots as shown in Figures 5a to 5c, and for categorical variable distribution, data analysis results can be provided as a distribution analysis result table and a bar graph as shown in Figure 6.

[0144] The values ​​calculated in the continuous variable distribution analysis result table are: (a) Total N: The total number of samples included in the data (b) Valid N(%): The number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Min~Max: Minimum to maximum value of the variable (e) Mean ± standard deviation: The mean ± standard deviation of the variable (f) Mean(95%CIs): Mean of the variable (95% confidence interval) (g)Median(IQR): The median (inter-quartile range) of the variable (h)Skewness: Skewness of the variable (i) Kurtosis: The kurtosis of the variable (j) Lilliefors test for normality, p value: The significance probability value (p value) of the normal distribution of the variable in question as a result of testing it with the Lilliefors method (k) Shapiro-Wilk test for normality, p value: p value of the normal distribution of the variable tested using the Shapiro-Wilk method The categorical variables provide data analysis results as distribution analysis result tables and bar plots.

[0145] The values ​​calculated in the categorical variable distribution analysis result table are: (a) Total N: The total number of samples included in the data (b) Valid N(%): The number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Subgroup: Name of the subgroup included in the categorical variable (e) N(%): The number and percentage of samples included in the subgroup (f) 95%CI: 95% confidence interval for the percentage of samples included in the subgroup It is.

[0146] [When all variables are individually measured] In the process of analyzing the relationship between variables, when all variables are single measurement variables, data analysis is performed by dividing the data into analysis of the relationship between two variables and analysis of the relationship between three or more variables. In the case of analysis of the relationship between two variables, data analysis is performed according to (C) the relationship between continuous variables and continuous variables, (D) the relationship between continuous variables and categorical variables, and (E) the relationship between categorical variables and categorical variables. In the case of analysis of the relationship between three or more variables, data analysis results are provided by dividing the data into when all variables are continuous variables, when all variables are categorical variables, and when continuous variables and categorical variables are mixed.

[0147] [Relationship between continuous variables] In the case of (C) the relationship between continuous variables, the method includes the steps of: conducting a normal distribution test for the two continuous variables; and judging whether the two variables all follow a normal distribution based on the result of the normal distribution test; if they do follow a normal distribution, performing data analysis by setting up "Pearson correlation analysis"; if they do not follow a normal distribution, performing data analysis by setting up "Spearman correlation analysis"; and providing the analysis results as an analysis result table and an analysis result graph.

[0148] The numerical values ​​presented in the analysis result table provided as the analysis result of the relationship between the continuous variables are: (a) Correlation coefficient (b) 95% confidence interval of the correlation coefficient (c) The significance probability value (p value) calculated as a result of testing whether the correlation coefficient is 0 It is.

[0149] The analysis result diagram provided as the analysis result of the relationship between the continuous variables may be a correlation scatter plot in which variables are set on the x-axis and y-axis and a regression curve is displayed, as shown in FIG. 9.

[0150] [Relationship between continuous and categorical variables] In the case of (D) the relationship between continuous variables and categorical variables, Data analysis is performed by dividing it into mean difference analysis, correlation analysis, and dichotomous reaction prediction performance analysis. In the case of the mean variance analysis, the process includes extracting the number of categorical variable subgroups (m), and performing a normal distribution test on the continuous variables when the number of categorical variable subgroups (m) is two and when the number of categorical variable subgroups is three or more, respectively, and determining whether the continuous variables follow a normal distribution when the number of categorical variable subgroups (m) is two. If the number of categorical variable subgroups (m) is two and the continuous variables do not follow a normal distribution, set the "Wilcoxon rank-sum test", and if the number of categorical variable subgroups (m) is two and the continuous variables follow a normal distribution, perform the "Levene's test" to test whether the variances of each subgroup are the same, set the "Student's Test" if the variances of each subgroup are the same, and if the variances of each subgroup are not the same, set the "Welch's T test" to perform data analysis. If the number of categorical variable subgroups (m) is three or more and the continuous variables follow a normal distribution, perform a one-way analysis of variance (1-way and if the continuous variables do not follow a normal distribution, then a Kruskal-Wallis H test is used to analyze the data.

[0151] The method further includes a step of performing a post-hoc analysis after the data analysis using the 1-way ANOVA and the Kruskal-Wallis H test, The algorithms used in the post-hoc validation performed after the data analysis by the 1-way ANOVA include the Bonferroni test, the Tukey test, the Scheffe test, and the Dunnett test, and the algorithms used in the post-hoc validation performed after the data analysis by the Kruskal-Wallis H test include the Bonferroni test, the FDR (False Discovery rate), and the Dunn's test.

[0152] The dichotomous response prediction analysis includes a process of extracting the number (m) of categorical variable subgroups, and when the number (m) of categorical variable subgroups is two, setting up a ROC curve analysis algorithm to control the data analysis.

[0153] The correlation analysis includes the steps of performing a normal distribution analysis on a continuous variable, determining whether the continuous variable follows a normal distribution, determining whether the categorical variable is ordinal if the continuous variable follows a normal distribution, setting a Polyserial correlation analysis to perform data analysis if the categorical variable is ordinal, and setting a Point polyserial correlation analysis to perform data analysis if the categorical variable is not ordinal, determining whether the categorical variable is ordinal if the continuous variable does not follow a normal distribution, and setting a Polychoric correlation analysis algorithm to perform data analysis if the categorical variable is an ordinal variable, and setting a Rank Polyserial correlation analysis algorithm to perform data analysis if the categorical variable is not an ordinal variable.

[0154] [Relationship between categorical variables and categorical variables] The (E) relationship between the categorical variable and the variable-type variable includes a step of classifying the number of subgroups (m, n) included in the categorical variable into two cases and other cases, and performing an independence test when the number of subgroups (m, n) included in the categorical variable is two, and performing an independence test, a trend test, and a correlation analysis when the number of subgroups (m, n) included in the categorical variable is not two, The independence test includes the steps of preparing a 2×2 cross table combining two subgroups of a categorical variable, preparing an expectation table under the assumption that the categorical variables are independent, and judging whether the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells; and, when the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells, taking into consideration the computing capacity, setting up a Fisher's exact test if the computing capacity is sufficient, and setting up a Chi-squared test with Yates correction if the computing capacity is insufficient; and when the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is less than 25% of the total number of cells, judging whether the cross table is in the form of 2×2, setting up a Chi-squared test with Yates correction, and when the cross table is not in the form of 2×2, setting up a Chi-squared test with Yates correction. and setting up a test and conducting data analysis.

[0155] The trend test includes the steps of creating an m×n cross table combining two subgroups of categorical variables and determining whether the conditions m≧3 and n≧3 are met, and selecting and setting a Linear by linear association test if the conditions m≧3 and n≧3 are met, and selecting and setting a Cochran's Q test if the conditions m≧3 and n≧3 are not met, thereby performing data analysis.

[0156] The correlation analysis includes a process of determining whether all categorical variables are ordinal, and if so, selecting and setting Polychoric correlation analysis; if not all categorical variables are ordinal, determining whether all categorical variables are nominal, and if so, selecting and setting Cramer's V analysis; if not, setting Rank polyserial correlation analysis to perform data analysis.

[0157] [Analysis of relationships between three or more all independently measured variables] (F) Analysis of the relationship between three or more variables is performed by dividing the data into three or more groups: when all of the groups are continuous, when all of the groups are categorical, and when there is a mixture of continuous and categorical variables. When all data are continuous, the process includes setting up PCA (Principle Component Analysis) and performing data analysis. When continuous and categorical variables are mixed, the method includes the steps of: using each of the continuous variables as a dependent variable and setting up univariable linear regression to analyze the individual influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and setting up multivariable linear regression to analyze the integrated influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and when there is a categorical variable among the remaining variables, setting up ANCOVA (Analysis of Covariance) to perform data analysis; and using each of the continuous variables as a dependent variable and when all the remaining variables are categorical variables, performing 2-way ANOVA to perform data analysis.

[0158] When all the variables are categorical, the process includes determining whether there are dichotomous variables; if there are dichotomous variables, using each of the dichotomous variables as a dependent variable, performing univariate binary logistic regression to analyze the individual influence of the remaining variables on the dichotomous variables; using each of the dichotomous variables as a dependent variable, performing multivariable binary logistic regression to analyze the integrated influence of the remaining variables on the dichotomous variables; constructing a dichotomous response prediction model; and controlling the verification analysis of the constructed dichotomous response prediction model. When all the variables are categorical, a process of determining whether there are dichotomous variables is included; if there are dichotomous variables, a process of using each of the dichotomous variables as a dependent variable, setting up a univariate binary logistic regression to analyze the individual influence of the remaining variables on the dichotomous variables, setting up a multivariate binary logistic regression to analyze the integrated influence of the remaining variables on the dichotomous variables; a process of constructing a dichotomous response prediction model; and a process of controlling a verification analysis for the constructed dichotomous response prediction model. When there are no dichotomous variables and all variables are ternary or higher, each of the variables is used as a dependent variable, and the ternary variables are assumed to be ordinal variables and the ternary variables are assumed to be nominal variables. Then, for each of the variables, an analysis of the individual influences and integrated influences of the remaining variables on the ternary ordinal variable and the ternary nominal variable is performed. This analysis includes the steps of performing univariable ordinal logistic regression to analyze the individual influences of the remaining variables on the ternary ordinal variable, performing multivariable ordinal logistic regression to analyze the integrated influence of the remaining variables on the ternary ordinal variable, performing univariable nominal logistic regression to analyze the individual influences of the remaining variables on the ternary nominal variable, and performing multivariable nominal logistic regression to analyze the integrated influence of the remaining variables on the ternary nominal variable.

[0159] The validation analysis of the dichotomous reaction prediction model is performed by a discrimination aspect prediction performance analysis, a calibration aspect prediction performance analysis, and a model performance cross-validation analysis.

[0160] The indicators used in the discrimination aspect prediction performance analysis are: (a) Performance analysis index (including 95% confidence interval) Area under the curve (95%CI) Sensitivity, Specificity ·PPV(positive predictive value), NPV(negative predictive value) ·ACC(accuracy), MIS(miss-classification rate) ·FPR(False Positive Rate), FNR(False Negative Rate), FDR(False Discovery Rate), FOR(False Omission Rate) ·LR+(Positive Likelihood Ratio), LR-(Negative Likelihood Ratio), DOR(Diagnostic Odds Ratio) (b) Visualization of performance analysis results ·ROC curve It is.

[0161] The indices used in the Calibration aspect performance analysis are: (a) Performance analysis index ·AIC(Akaike Information Criterion) ·BIC(Bayes Information Criterion) ·Nagelkerke R2 ·Hosmer-Lemeshow test P value Brier score ·Spigelhalter Z score with P value ·Linear regression line in calibration plot Intercept, 95% confidence interval and p-value Slope, 95% confidence interval and p-value (b) Visualization of performance analysis results Calibration plot Decile plot Calibration belt It is.

[0162] The model performance cross-validation analysis includes: (a) Methodology used for performance cross-validation ·LOOCV(leave-one-out cross-validation) K-fold CV (cross validation) Permutation test Bootstrapping (b) Visualization of cross-validation results ·ROC curve Calibration plot Decile plot Calibration belt It is.

[0163] [Analysis when all variables are repeated measurements] (G) If all variables are repeated measurements, the data will be analyzed by dividing the data into those with two repeated measurements and those with three or more repeated measurements, and those with continuous variables and those with categorical variables. When there are two repeated measurement variables and the repeated measurement variables are continuous variables, the method includes the steps of: performing normal distribution analysis on the continuous variables to determine whether the continuous variables follow a normal distribution; if they do follow a normal distribution, setting up a paired sample T test; if they do not follow a normal distribution, setting up a Wilcoxon signed-rank test to perform data analysis.

[0164] In addition, when there are two repeatedly measured variables and the repeatedly measured variables are continuous variables, the method may further include the step of setting up an ICC (Intraclass Correlation Coefficient) analysis and performing data analysis. When there are two repeatedly measured variables and the repeatedly measured variables are categorical variables, the method includes the steps of: performing data analysis using McNemar's test and Cohens' Kappa when the number of subgroups (m, n) contained in the categorical variables is two (m=2, n=2); performing data analysis using Cochran-Armitage test for trend when the number of subgroups (m, n) contained in the categorical variables is two or more (m=2, n≧3); and performing data analysis using McNemar-Bowker test when the number of subgroups (m, n) contained in the categorical variables is three or more (m≧3, n≧3).

[0165] When there are three or more repeated measurement variables and the repeated measurement variables are continuous variables, the process includes setting up Linear mixed effect model analysis, GEE (Generalized Estimating Equation) analysis, and Repeated measures 1-way ANOVA to perform data analysis.

[0166] When there are three or more repeatedly measured variables and the repeatedly measured variables are categorical variables, the method includes a process of setting up a Generalized Mixed Effect Model Analysis (GEE) analysis to perform data analysis.

[0167] [Analysis when single measured variables are mixed] (H) When a single measurement variable and a repeatedly measured variable are mixed, the repeatedly measured variables are classified into those in which the repeatedly measured variables are continuous variables and those in which the repeatedly measured variables are categorical variables. When the repeated measurement variables are continuous variables, data analysis is performed by setting up a Linear mixed effect model analysis or GEE (Generalized Estimating Equation) analysis, and when the repeated measurement variables are continuous variables and the single measurement variable is a categorical variable, data analysis is performed by setting up a Repeated measures 2-way ANOVA.

[0168] When the repeated measurement variables are categorical variables, the method includes a process of setting up a generalized mixed effect model analysis (GEE) analysis to perform data analysis.

[0169] [Analysis of Survival Materials] (I) Analysis of survival data is divided into cases where there are only survival times and event occurrence variables, and cases where there are survival times, event occurrence variables, and single measurement data. When only survival time and event occurrence variables are available, a Kaplan-Meier curve analysis algorithm is set up to perform data analysis; when survival time, event occurrence data, and single measurement data are available, the method includes setting up univariable cox proportional hazards regression to analyze the individual influence of single measurement variables on survival, setting up multivariable cox proportional hazards regression to analyze the integrated influence of single measurement variables on survival, and setting up Kaplan-Meier curve analysis to perform log rank (comparison of survival probability differences between detailed categories).

[0170] The method may further include a process of constructing a survival probability prediction model, and a process of analyzing the survival probability prediction model by performing a discrimination aspect prediction performance analysis at time=t, a calibration aspect prediction performance analysis at time=t, and a survival probability prediction model cross-validation analysis.

[0171] The indicators used in the discrimination aspect prediction performance analysis at Time=t are, as described above, The indicators used in the Discrimination aspect performance analysis are: (a) Performance analysis index (including 95% confidence interval) Area under the curve (95%CI) Sensitivity, Specificity ·PPV(positive predictive value), NPV(negative predictive value) ·ACC(accuracy), MIS(miss-classification rate) ·FPR(False Positive Rate), FNR(False Negative Rate), FDR(False Discovery Rate), FOR(False Omission Rate) ·LR+(Positive Likelihood Ratio), LR-(Negative Likelihood Ratio), DOR(Diagnostic Odds Ratio) (b) Visualization of performance analysis results ·ROC curve It is.

[0172] The indices used in the Calibration aspect performance analysis are: (a) Performance analysis index ·AIC(Akaike Information Criterion) ·BIC(Bayes Information Criterion) ·Nagelkerke R2 ·Hosmer-Lemeshow test P value Brier score ·Spigelhalter Z score with P value ·Linear regression line in calibration plot Intercept, 95% confidence interval and p-value Slope, 95% confidence interval and p-value (b) Visualization of performance analysis results Calibration plot Decile plot Calibration belt It is.

[0173] In the survival probability prediction model performance cross-validation, (a) Methodology used for performance cross-validation ·LOOCV(leave-one-out cross-validation) K-fold CV (cross validation) Permutation test Bootstrapping (b) Visualization of cross-validation results ·ROC curve Calibration plot Decile plot Calibration belt It is. [Industrial Applicability]

[0174] The present invention enables data analysis by selecting an optimal algorithm based on variable attributes, and can provide an automated data analysis system that is adaptive to the increasing diversity of variable data as clinical processes and fields are subdivided into various types. Therefore, this technology can be widely used in the medical, bio, and statistical analysis industries to realize its practical and economical value.

Claims

1. The data analysis system includes a variable attribute definition means (10) for extracting characteristics of variables included in the data, classifying the variables according to the form and characteristics of the constituent values ​​of the variables, and defining the attributes of the classified variables; an algorithm management means (20) for storing and managing data analysis algorithms and providing the algorithms in response to a request from a data analysis control means (30); a data analysis control means (30) for defining attributes of variables included in the data through the variable attribute definition means (10), selecting an algorithm for data analysis based on the variable attributes, and controlling the execution of data analysis by the data analysis means (40); and a data analysis means (40) for performing data analysis according to the algorithm set by the data analysis control means (30), and providing the result information as a distribution analysis result table and a distribution analysis result diagram, The variable attribute definition means (10) classifies the data into continuous variables and categorical variables according to the form of the constituent values ​​of the variables included in the data, and classifies the categorical variables into a first classification variable, which is classified into an ordinal variable and a nominal variable, and a second classification variable including a single measured variable, a repeatedly measured variable, and survival data according to the characteristics of the variables included in the data, and defines the attributes thereof; The data analysis control means (30) comprises: a variable relation analysis control means (31) for performing a variable relation analysis on the automatically classified variables according to a combination of their types by referring to the variable attributes, and judging whether the analysis is a variable per se distribution analysis or a relation analysis between two or more variables; a variable per se distribution analysis control means (32) for analyzing and controlling data for each of continuous variables and categorical variables when the analysis is a variable per se distribution analysis based on a result of the variable relation analysis control means (31); and a variable relation analysis control means (33) for analyzing and controlling data by setting an algorithm by classifying when all variables are single-measured variables, when all variables are repeatedly measured variables, when single-measured variables and repeatedly measured variables are mixed, or when the data is survival data when the analysis is a relation analysis between two or more variables based on a result of the variable relation analysis control means (31).

2. The data analysis control means (30) further includes a normal distribution test means (34) for continuous variables, which calculates a significance probability value (p-value; p1) by performing a Lilliefors test for the continuous variables, calculates a significance probability value (p-value; p2) by performing a Shapiro-Wilk test, and calculates a normal distribution test result for the continuous variables by comparing p1 and p2 with a reference value α; 2. The system for automating exploratory data analysis based on variable attributes according to claim 1, wherein the normal distribution test result for a continuous variable by the normal distribution test means (34) is determined to not follow a normal distribution if the condition p1<α (AND or OR) p2<α is satisfied, and is determined to follow a normal distribution if the condition is not satisfied.

3. 3. The automated system for exploratory data analysis based on variable attributes according to claim 2, wherein the normal distribution test means for continuous variables (34) further includes a standard value setting means (34a) for providing a standard value setting process so that a user can set a standard value (α).

4. The values ​​calculated in the distribution analysis result table of the continuous variable are: (a) Total N: The total number of samples included in the data (b) Valid N (%): Number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Min to Max: the minimum value to the maximum value of the variable (e) Mean ± standard deviation: the mean ± standard deviation of the variable (f) Mean (95% CIs): Mean (95% confidence interval) of the variable (g) Median (IQR): Inter-quartile range of the variable (h) Skewness: Skewness of the variable (i) Kurtosis: The kurtosis of the variable (j) Lilliefors test for normality, p value: significance probability value (p value) based on the result of testing the normal distribution of the variable using the Lilliefors method (k) Shapiro-Wilk test for normality, p value: p value based on the results of testing the normal distribution of the variable using the Shapiro-Wilk method The values ​​calculated in the distribution analysis result table of the categorical variable are: (a) Total N: The total number of samples included in the data (b) Valid N (%): Number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Subgroup: Name of a subgroup included in a categorical variable (e) N (%): The number and percentage of samples included in the subgroup 3. The variable attribute-based exploratory data analysis automation system according to claim 1 or 2, characterized in that it comprises: (f) 95% CI: a 95% confidence interval of the percentage of samples included in the subgroup.

5. The variable relationship analysis control means (33) Provide a process for controlling data analysis by setting algorithms according to the following: when all variables are single measured, when all variables are repeated measured, when single and repeated measured variables are mixed, and when data is survival data; [When all variables are individually measured] 3. The system for automating exploratory data analysis based on variable attributes according to claim 1 or 2, characterized in that it comprises a process for performing data analysis by dividing into a relationship analysis between two variables and a relationship analysis between three or more variables when all variables are solely measured variables, and performing data analysis according to a relationship between a continuous variable and a continuous variable, a relationship between a continuous variable and a categorical variable, and a relationship between a categorical variable and a categorical variable when the relationship analysis is between three or more variables, and performing data analysis by dividing into a relationship between all variables, a relationship between all variables and a categorical variable, and a relationship between three or more variables when the relationship analysis is between three or more variables.

6. [Relationship between continuous variables] In the case of the relationship between continuous variables, the data analysis control process includes the steps of: performing a normal distribution test on the two continuous variables; judging whether both variables follow normal distribution based on the result of the normal distribution test; if they follow normal distribution, performing data analysis control by setting Pearson correlation analysis; if they do not follow normal distribution, performing data analysis control by setting Spearman correlation analysis; and providing the analysis results as an analysis result table and an analysis result chart; [Relationship between continuous and categorical variables] In the case of a relationship between a continuous variable and a categorical variable, the data analysis control process Data analysis is performed by dividing it into mean difference analysis, correlation analysis, and dichotomous response prediction performance analysis. The mean variance analysis includes the steps of extracting the number of categorical variable subgroups (m), and performing a normal distribution test on the continuous variables when the number of the extracted categorical variable subgroups (m) is two and when the number of the categorical variable subgroups is three or more, respectively, and determining whether the continuous variables follow a normal distribution when the number of the categorical variable subgroups (m) is two. If the number of the categorical variable subgroups (m) is two and the continuous variables do not follow a normal distribution, a Wilcoxon rank-sum test is set, and if the number of the categorical variable subgroups (m) is two and the continuous variables follow a normal distribution, a Levene's test is performed to test whether the variances of the subgroups are the same. If the variances of the subgroups are the same, a "Student's Test" is set, and if the variances of the subgroups are not the same, a "Welch's Test" is performed. and when the number of categorical variable subgroups (m) is 3 or more, if the continuous variables follow a normal distribution, setting a one-way analysis of variance (1-way ANOVA) to control the data analysis, and if the continuous variables do not follow a normal distribution, setting a Kruskal-Wallis H test to control the data analysis, The correlation analysis includes the steps of: performing a normal distribution analysis on a continuous variable; determining whether the continuous variable follows a normal distribution; determining whether the categorical variable is ordinal if the continuous variable follows a normal distribution; setting a polyserial correlation analysis to control the data analysis if the categorical variable is ordinal, and setting a point polyserial correlation analysis to control the data analysis if the categorical variable is not ordinal; determining whether the categorical variable is ordinal if the continuous variable does not follow a normal distribution; setting a polychoric correlation analysis to control the data analysis if the categorical variable is ordinal, and setting a rank polyserial correlation analysis to control the data analysis if the categorical variable is not ordinal; The dichotomous response prediction performance analysis includes the steps of: extracting a number (m) of categorical variable subgroups; and, when the number (m) of extracted categorical variable subgroups is two, setting an ROC curve analysis algorithm to control data analysis; [Relationship between categorical variables] In the case of the relationship between categorical variables, the data analysis control process includes a step of controlling the data analysis by dividing the number of subgroups (m, n) included in the categorical variables into two cases and other cases, and dividing the number of subgroups (m, n) included in the categorical variables into two cases, namely, an independence test, and a correlation analysis, if the number of subgroups (m, n) included in the categorical variables is not two. The independence test includes the steps of: preparing a 2×2 cross table combining subgroups of two categorical variables; preparing an expectation table under the assumption that the categorical variables are independent; and judging whether the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variables is 25% or more of the total number of cells; and, if the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variables is 25% or more of the total number of cells, taking into consideration the computational capacity, if the computational capacity is sufficient, setting up Fisher's exact test, or if the computational capacity is not sufficient, setting up Chi-squared test with Yates' and if the number of cells with an expected value < 5 in an m x n cross table of the number of subgroups (m, n) included in the categorical variable is less than 25% of the total number of cells, determining whether the cross table is in a 2 x 2 format, setting a Chi-squared test with Yates correction if the cross table is in a 2 x 2 format, and setting a Chi-squared test if the cross table is not in a 2 x 2 format to control the data analysis; The trend test includes a process of preparing an m×n cross table by combining two subgroups of categorical variables, and judging whether the condition m≧3 and n≧3 is satisfied; and a process of selecting and setting a Linear by Linear Association test if the condition m≧3 and n≧3 is satisfied, and a process of setting a Cochran's Q test if the condition m≧3 and n≧3 is not satisfied to control data analysis. The correlation analysis includes a process of judging whether all categorical variables are ordinal, and if all are ordinal, selecting and setting Polychoric correlation analysis, and if all are not ordinal, judging whether all are nominal variables, and if all are nominal, selecting and setting Cramer's V analysis, and if all are not nominal, selecting Rank Polyserial Correlation Analysis. setting up the analysis to control the data analysis; [Analysis of Relationships Between Three or More Total Single Measured Variables] The relationship analysis data analysis control process among all three or more independent measurement variables includes: When three or more variable groups are all continuous, when they are all categorical, or when there is a mixture of continuous and categorical variables, data analysis is performed by dividing them into categories. When all are continuous, it comprises the step of setting up PCA (Principle Component Analysis) and controlling the data analysis; When continuous variables and categorical variables are mixed, the method includes the steps of: using each of the continuous variables as a dependent variable and setting up a uniform linear regression to analyze the individual influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and setting up a multivariable linear regression to analyze the integrated influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and when there is a categorical variable among the remaining variables, setting up ANCOVA (Analysis of Covariance) to perform data analysis control; and using each of the continuous variables as a dependent variable and when all the remaining variables are categorical variables, performing 2-way ANOVA to perform data analysis control. a process of determining whether there is a dichotomous variable when all the variables are categorical variables; a process of using each of the dichotomous variables as a dependent variable when there are dichotomous variables, and setting up a univariable binary logistic regression to analyze the individual influence of the remaining variables on the dichotomous variables; a process of using each of the dichotomous variables as a dependent variable, and when there is a dichotomous variable among the remaining variables, using each of the dichotomous variables as a dependent variable to construct a dichotomous response prediction model; and a process of controlling a verification analysis for the constructed dichotomous response prediction model. a process of determining whether there is a dichotomous variable when all the variables are categorical variables; a process of setting up a univariable binary logistic regression to analyze the individual influence of the remaining variables on the dichotomous variable when the dichotomous variables are present, setting up a multivariable binary logistic regression to analyze the integrated influence of the remaining variables on the dichotomous variable; a process of constructing a dichotomous response prediction model; and a process of controlling a verification analysis for the constructed dichotomous response prediction model. When there are no dichotomous variables and all variables are ternary or higher, each of the variables is used as a dependent variable, and the ternary variables are assumed to be ordinal variables. When the ternary variables are assumed to be nominal variables, the remaining variables are analyzed for their individual and integrated influences on the ternary ordinal and ternary nominal variables. The following analyses are performed: a process of setting a uniform ordinal logistic regression to analyze the individual influences of the remaining variables on the ternary ordinal variables; a process of setting a multi-ordinal logistic regression to analyze the integrated influences of the remaining variables on the ternary ordinal variables; a process of setting a uniform nominal logistic regression to analyze the individual influences of the remaining variables on the ternary nominal variables; and a process of setting a multi-ordinal logistic regression to analyze the individual influences of the remaining variables on the ternary ordinal variables. and a step of setting a regression to analyze the overall influence of the remaining variables on the three-choice nominal variable.

7. The numerical values ​​presented in the analysis result table provided as the analysis result of the relationship between the continuous variables are: (a) Correlation coefficient (b) 95% confidence interval of correlation coefficient 7. The variable attribute-based exploratory data analysis automation system according to claim 6, further comprising: (c) a significance probability value (p-value) calculated as a result of testing whether the correlation coefficient is 0.

8. The analysis result diagram provided as the analysis result of the relationship between the continuous variables is a correlation scatter diagram in which x-axis and y-axis variables are set and a regression curve is displayed.

7. The variable attribute-based exploratory data analysis automation system according to claim 6, further comprising a scatter plot.

9. In the case of a relationship between a continuous variable and a categorical variable, the data analysis control process The method further includes a step of performing a post-hoc analysis after the data analysis using the 1-way ANOVA and the Kruskal-Wallis H test, 7. The system for automating exploratory data analysis based on variable attributes according to claim 6, wherein the algorithms used in the post-hoc validation performed after the data analysis by the 1-way ANOVA include a Bonferroni test, a Tukey test, a Scheffe test, and a Dunnett test, and the algorithms used in the post-hoc validation after the data analysis by the Kruskal-Wallis H test include a Bonferroni test, an FDR (False Discovery rate), and a Dunn's test.

10. The validation analysis of the dichotomous reaction prediction model was performed by a discrimination aspect prediction performance analysis, a calibration aspect prediction performance analysis, and a model performance cross-validation analysis. The indicators used in the Discriminatory aspect prediction performance analysis are: (a) Performance analysis index (including 95% confidence interval) ・AUC (95% CI) ・Sensitivity, Specificity ・PPV (positive predictive value), NPV (nega tive predictive value) ・ACC (accuracy), MIS (miss-classification rate) ・FPR (False Positive Rate), FNR (False Negative Rate), FDR (False Discovery Rate), FOR (False Mission Rate) ・LR+ (Positive Likelihood Ratio), LR- (Negative Likelihood Ratio), DOR (Diagnostic Odds Ratio) (b) Visualization of performance analysis results ・ROC curve The indices used in the Calibration aspect prediction performance analysis are: (a) Performance analysis index ・AIC (Akaike Information Criterion) ・BIC (Bayes Information Criterion) ・Nagelkerke R2 ・Hosmer-Lemeshow test P value ・Brier score ・Spigelhalter Z score with P value Linear regression line in calibration plot Intercept, 95% confidence interval and p-value Slope, 95% confidence interval and p-value (b) Visualization of performance analysis results ・Calibration plot ・Decile plot ・Calibration belt The model performance cross-validation analysis includes: (a) Methods used for performance cross-validation ・LOOCV (leave-one-out cross-validation) ・K-fold CV (cross validation) ・Permutation test ・Bootstrapping (b) Visualization of cross-validation results ・ROC curve ・Calibration plot ・Decile plot ・Calibration belt 7. The variable attribute-based exploratory data analysis automation system according to claim 6, comprising:

11. [Analysis when all variables are repeated measurements] The whole repeated measurement variable analysis data analysis control process includes: Data analysis is performed by dividing the data into cases where there are two repeatedly measured variables and cases where there are three or more repeatedly measured variables, and cases where the repeatedly measured variables are continuous variables and cases where the repeatedly measured variables are categorical variables. When the number of repeated measurement variables is two and the repeated measurement variables are continuous variables, the method includes the steps of: performing a normal distribution analysis on the continuous variables to determine whether the continuous variables follow a normal distribution; if they do follow a normal distribution, setting a paired sample T test; and if they do not follow a normal distribution, setting a Wilcoxon signed-rank test to control the data analysis; In the case where there are two repeated measurement variables and the repeated measurement variables are categorical variables, the method includes the steps of: when the number of subgroups (m, n) included in the categorical variables is two (m=2, n=2), setting McNemar's test and Cohens' Kappa to control the data analysis; when the number of subgroups (m, n) included in the categorical variables is two, three or more (m=2, n≧3), setting Cochran-Armitage test for trend to control the data analysis; and when the number of subgroups (m, n) included in the categorical variables is three or more (m≧3, n≧3), setting McNemar-Bowker test to control the data analysis; When the number of repeated measurement variables is three or more and the repeated measurement variables are continuous variables, the method includes a step of setting up a Linear mixed effect model analysis, a Generalized Estimating Equation (GEE) analysis, or a Repeated measures 1-way ANOVA to control the data analysis; 3. The system for automating exploratory data analysis based on variable attributes according to claim 1 or 2, further comprising a step of setting up Generalized mixed effect model analysis (GEE (Generalized Estimating Equation) analysis) to control data analysis when there are three or more repeated measurement variables and the repeated measurement variables are categorical variables.

12. 12. The system for automating exploratory data analysis based on variable attributes according to claim 11, further comprising the step of setting an ICC (Intraclass Correlation Coefficient) analysis to control data analysis when there are two repeated measurement variables and the repeated measurement variables are continuous variables.

13. [Analysis when single measurement variables are mixed] The data analysis control process when the single measurement variables are mixed is as follows: The repeated measurement variables are classified into two types: continuous variables and categorical variables. When the repeated measurement variables are continuous variables, a Linear mixed effect model analysis or a Generalized Estimating Equation (GEE) analysis is set up to control the data analysis, and when the repeated measurement variables are continuous variables and the single measurement variable is a categorical variable, a Repeated measures 2-way ANOVA is set up to control the data analysis; 3. The system for automating exploratory data analysis based on variable attributes according to claim 1 or 2, further comprising a step of setting up Generalized mixed effect model analysis (GEE (Generalized Estimating Equation) analysis) to control data analysis when the repeated measurement variables are categorical variables.

14. [Analysis of Survival Materials] The survival data analysis control process includes: The cases are divided into those with only survival time and event occurrence variables and those with survival time, event occurrence variables, and single measurement data.

3. The system for automating exploratory data analysis based on variable attributes according to claim 1, further comprising: a step of setting a Kaplan-Meier curve analysis to control data analysis when only survival time and event occurrence variables are present; and a step of setting a Univariable Cox proportional hazards regression to control an analysis of the individual influence of a single measurement variable on survival when survival time, event occurrence data, and single measurement data are present; a step of setting a Multivariable Cox proportional hazards regression to analyze and control the integrated influence of a single measurement variable on survival; and a step of setting a Kaplan-Meier curve analysis to control a Log rank test (comparison of survival probability difference between detailed categories).

15. The survival data analysis control process includes: The system for automating exploratory data analysis based on variable attributes according to claim 14, further comprising the steps of: constructing a survival probability prediction model; and analyzing and controlling the survival probability prediction model by performing a discrimination aspect prediction performance analysis at time=t, a calibration aspect prediction performance analysis at time=t, and a survival probability prediction model cross-validation analysis.

16. A method for automating data analysis, comprising: a variable attribute definition step of defining variable attributes for automating data analysis by a computer; and a step of selecting and setting an algorithm for data analysis by the computer based on the variable attributes, The step of defining one variable attribute includes: According to the form of the constituent values ​​of the variables included in the data, the data are classified into continuous variables and categorical variables, and the categorical variables are classified into a first classification variable including an ordinal variable and a nominal variable, and a second classification variable including a single measured variable, a repeatedly measured variable, and survival data according to the characteristics of the variables included in the data, and the attributes of the second classification variable are defined; The data analysis process involves selecting algorithms for data analysis based on variable attributes and performing data analysis. A variable relationship analysis process in which a variable relationship analysis is performed on the automatically classified variables according to the combination of their types to determine whether it is a distribution analysis of the variable itself or a relationship analysis between two or more variables; a variable distribution analysis step for analyzing the distribution of continuous variables and categorical variables based on the variable relationship analysis result of the variable relationship analysis step, and providing the results as a distribution analysis result table and a distribution analysis result chart; and a variable relationship analysis process for selecting and setting an algorithm according to whether the variables are all individually measured variables, all are repeatedly measured variables, there is a mixture of individually measured variables and repeatedly measured variables, or the variables are survival data, in the case of a relationship analysis between two or more variables based on the result of the variable relationship analysis process, and providing a data analysis result.

17. The variable distribution analysis step further includes a continuous variable normal distribution test step including: calculating a significance probability value (p value: p1) by performing a Lilliefors test; calculating a significance probability value (p value: p2) by performing a Shapiro-Wilk test; and calculating a normal distribution test result for the continuous variable by comparing a set reference value (α) with p1 and p2; 17. The method for automating exploratory data analysis based on variable attributes according to claim 16, wherein the normal distribution test result for the continuous variable is determined to not follow a normal distribution if the condition p1<α (AND or OR) p2<α is satisfied, and is determined to follow a normal distribution if the condition is not satisfied.

18. In the process of analyzing the distribution of the variables themselves, The values ​​calculated in the continuous variable distribution analysis result table are: (a) Total N: The total number of samples included in the data (b) Valid N (%): Number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Min to Max: the minimum value to the maximum value of the variable (e) Mean ± standard deviation: the mean ± standard deviation of the variable (f) Mean (95% CIs): Mean (95% confidence interval) of the variable (g) Median (IQR): Inter-quartile range of the variable (h) Skewness: Skewness of the variable (i) Kurtosis: The kurtosis of the variable (j) Lilliefors test for normality, p value: significance probability value (p value) based on the result of testing the normal distribution of the variable using the Lilliefors method (k) Shapiro-Wilk test for normality, p value: p value based on the results of testing the normal distribution of the variable using the Shapiro-Wilk method The values ​​calculated in the categorical variable distribution analysis result table are: (a) Total N: The total number of samples included in the data (b) Valid N (%): Number and percentage of samples excluding missing values (c) Missing N(%): The number and percentage of samples that are missing (d) Subgroup: Name of a subgroup included in a categorical variable (e) N (%): The number and percentage of samples included in the subgroup 18. The method for automating exploratory data analysis based on variable attributes according to claim 16 or 17, characterized in that it includes (f) 95% CI: a 95% confidence interval for the percentage of samples included in the subgroup.

19. [When all variables are individually measured] In the above intervariable relationship analysis process, if all variables are single measurement variables, a relationship analysis between two variables is performed.

18. The method for automating exploratory data analysis based on variable attributes according to claim 16 or 17, characterized in that the method comprises the steps of: performing data analysis by classifying relationship analysis between three or more variables; in the case of relationship analysis between two variables, performing data analysis according to (C) the relationship between continuous variables and continuous variables, (D) the relationship between continuous variables and categorical variables, and (E) the relationship between categorical variables and categorical variables; and (F) in the case of relationship analysis between three or more variables, performing data analysis by classifying when all variables are continuous variables, when all variables are categorical variables, and when continuous variables and categorical variables are mixed.

20. [Relationship between continuous variables] (C) in the case of a relationship between continuous variables, performing a normal distribution test on the two continuous variables; and judging whether the two variables all follow a normal distribution based on the result of the normal distribution test, and if they do follow a normal distribution, performing data analysis by setting "Pearson correlation analysis", and if they do not follow a normal distribution, performing data analysis by setting "Spearman correlation analysis", and providing the analysis results as an analysis result table and an analysis result chart; [Relationship between continuous and categorical variables] In the case of the relationship between a continuous variable and a categorical variable (D), Data analysis is performed by dividing it into mean difference analysis, correlation analysis, and dichotomous response prediction performance analysis. The mean variance analysis includes the steps of extracting the number of categorical variable subgroups (m), and performing a normal distribution test on the continuous variables when the number of the extracted categorical variable subgroups (m) is two and when the number of the categorical variable subgroups is three or more, respectively, and determining whether the continuous variables follow a normal distribution when the number of the categorical variable subgroups (m) is two, and performing a "Wilcoxon rank-sum test" when the continuous variables do not follow a normal distribution when the number of the categorical variable subgroups (m) is two, and performing a "Levene's test" to test whether the variances of the subgroups are the same when the variances of the subgroups are the same, performing a "Student's Test" when the variances of the subgroups are the same, and performing a "Welch's T test" when the variances of the subgroups are not the same, to perform data analysis; When the number of categorical variable subgroups (m) is 3 or more, if the continuous variables follow a normal distribution, a one-way analysis of variance (1-way ANOVA) is set up to perform data analysis, and if the continuous variables do not follow a normal distribution, a Kruskal-Wallis H test is set up to perform data analysis; The correlation analysis includes the steps of: performing a normal distribution analysis on a continuous variable; determining whether the continuous variable follows a normal distribution; determining whether the categorical variable is ordinal if the continuous variable follows a normal distribution; setting a polyserial correlation analysis to perform data analysis if the categorical variable is ordinal, and setting a point polyserial correlation analysis to perform data analysis if the categorical variable is not ordinal; determining whether the categorical variable is ordinal if the continuous variable does not follow a normal distribution; setting a polychoric correlation analysis algorithm to perform data analysis if the categorical variable is ordinal, and setting a rank polyserial correlation analysis algorithm to perform data analysis if the categorical variable is not ordinal; The dichotomous response prediction analysis includes the steps of: extracting a number (m) of categorical variable subgroups; and, when the number (m) of the extracted categorical variable subgroups is two, setting an ROC curve analysis algorithm to control data analysis; [Relationship between categorical variables] The (E) relationship between the categorical variable and the variable-type variable includes a step of classifying the number of subgroups (m, n) included in the categorical variable into two cases and other cases, and performing an independence test when the number of subgroups (m, n) included in the categorical variable is two, and performing an independence test, a trend test, and a correlation analysis when the number of subgroups (m, n) included in the categorical variable is not two, The independence test includes the steps of: preparing a 2×2 cross table combining two subgroups of a categorical variable; preparing an expectation table under the assumption that the categorical variables are independent; and judging whether the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells; and, if the number of cells with an expectation value <5 in the m×n cross table of the number of subgroups (m, n) included in the categorical variable is 25% or more of the total number of cells, taking into consideration the computational capacity, if the computational capacity is sufficient, setting up a Fisher's exact test, and if the computational capacity is not sufficient, setting up a Chi-squared test with Yates' and if the number of cells with an expected value < 5 in an m x n cross table of the number of subgroups (m, n) included in the categorical variable is less than 25% of the total number of cells, determining whether the cross table is in a 2 x 2 format, and if the cross table is in a 2 x 2 format, setting a Chi-squared test with Yates correction, and if the cross table is not in a 2 x 2 format, setting a Chi-squared test to perform data analysis; The trend test includes the steps of: preparing an m×n cross table by combining two subgroups of categorical variables; determining whether the condition m≧3 and n≧3 is satisfied; and selecting and setting a Linear by Linear Association test if the condition m≧3 and n≧3 is satisfied; and selecting and setting a Cochran's Q test if the condition m≧3 and n≧3 is not satisfied, to perform data analysis; The correlation analysis includes a step of determining whether all categorical variables are ordinal, and if so, selecting and setting Polychoric correlation analysis; if not all categorical variables are ordinal, determining whether all categorical variables are nominal, and if so, selecting and setting Cramer's V analysis; if not all categorical variables are nominal, selecting and setting Rank polyserial correlation analysis to perform data analysis; [Analysis of Relationships Between Three or More Total Single Measured Variables] (F) Analysis of the relationship between three or more variables is performed by dividing the data into three or more groups: when all of the groups are continuous, when all of the groups are categorical, and when there is a mixture of continuous and categorical variables. When all the data are continuous, a PCA (Principle Component Analysis) is set up and data analysis is performed; When continuous variables and categorical variables are mixed, the method includes the steps of: using each of the continuous variables as a dependent variable and setting up a uniform linear regression to analyze the individual influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and setting up a multivariable linear regression to analyze the integrated influence of the remaining variables on the continuous variables; using each of the continuous variables as a dependent variable and when there is a categorical variable among the remaining variables, setting up ANCOVA (Analysis of Covariance) to perform data analysis; and using each of the continuous variables as a dependent variable and when all the remaining variables are categorical variables, performing 2-way ANOVA to analyze the data. a process of determining whether there is a dichotomous variable when all the variables are categorical variables; a process of analyzing the individual influence of the remaining variables on the dichotomous variables by performing univariable binary logistic regression using each of the dichotomous variables as a dependent variable when there are dichotomous variables; a process of analyzing the integrated influence of the remaining variables on the dichotomous variables by performing multivariable binary logistic regression using each of the dichotomous variables as a dependent variable; a process of constructing a dichotomous response prediction model; and a process of controlling a verification analysis for the constructed dichotomous response prediction model. When there are no dichotomous variables and all variables are ternary or higher, each of the variables is used as a dependent variable, and the ternary variables are assumed to be ordinal variables and the ternary variables are assumed to be nominal variables. Then, for each of the variables, an analysis of the individual influences and integrated influences of the remaining variables on the ternary ordinal variable and the ternary nominal variable is performed. The process of performing univariable ordinal logistic regression and analyzing the individual influences of the remaining variables on the ternary ordinal variable, the process of performing multivariable ordinal logistic regression and analyzing the integrated influences of the remaining variables on the ternary ordinal variable, the process of performing univariable nominal logistic regression and analyzing the individual influences of the remaining variables on the ternary ordinal variable, the process of performing multivariable nominal logistic regression and analyzing the individual influences of the remaining variables on the ternary nominal variable, and the process of performing multivariable nominal logistic regression and analyzing the individual influences of the remaining variables on the ternary nominal variable.

20. The method for automating exploratory data analysis based on variable attributes according to claim 19, further comprising the step of: performing regression to analyze the overall influence of the remaining variables on the three-choice nominal variables.

21. The numerical values ​​presented in the analysis result table provided as the analysis result of the relationship between the continuous variables are: (a) Correlation coefficient (b) 95% confidence interval of correlation coefficient The method for automating exploratory data analysis based on variable attributes according to claim 20, further comprising (c) a significance probability value (p-value) calculated as a result of testing whether the correlation coefficient is 0.

22. In the relationship analysis between the continuous variable and the categorical variable, The method further includes a step of performing a post-hoc analysis after the data analysis using the 1-way ANOVA and the Kruskal-Wallis H test, 21. The method for automating exploratory data analysis based on variable attributes according to claim 20, wherein the algorithms used in the post-hoc verification performed after the data analysis by the 1-way ANOVA include a Bonferroni test, a Tukey test, a Scheffe test, and a Dunnett test, and the algorithms used in the post-hoc verification after the data analysis by the Kruskal-Wallis H test include a Bonferroni test, an FDR (False Discovery rate), and a Dunn's test.

23. 21. The method for automating data analysis based on variable attributes according to claim 20, further comprising the step of setting up ROC curve analysis to control data analysis when the number of categorical variable subgroups (m) is two in the mean difference analysis.

24. In the process of controlling the validation analysis for the dichotomous reaction prediction model, the validation analysis for the dichotomous reaction prediction model is performed by a discrimination aspect prediction performance analysis, a calibration aspect prediction performance analysis, and a model performance cross-validation analysis; The indicators used in the discrimination aspect prediction performance analysis are: (a) Performance analysis index (including 95% confidence interval) ・AUC (95% CI) ・Sensitivity, Specificity ・PPV (positive predictive value), NPV (nega tive predictive value) ・ACC (accuracy), MIS (miss-classification rate) ・FPR (False Positive Rate), FNR (False Negative Rate), FDR (False Discovery Rate), FOR (False Mission Rate) ・LR+ (Positive Likelihood Ratio), LR- (Negative Likelihood Ratio), DOR (Diagnostic Odds Ratio) (b) Visualization of performance analysis results ・ROC curve The indices used in the Calibration aspect performance analysis are: (a) Performance analysis index ・AIC (Akaike Information Criterion) ・BIC (Bayes Information Criterion) ・Nagelkerke R2 ・Hosmer-Lemeshow test P value ・Brier score ・Spigelhalter Z score with P value Linear regression line in calibration plot Intercept, 95% confidence interval and p-value Slope, 95% confidence interval and p-value (b) Visualization of performance analysis results ・Calibration plot ・Decile plot ・Calibration belt The model performance cross-validation analysis includes: (a) Methods used for performance cross-validation ・LOOCV (leave-one-out cross-validation) ・K-fold CV (cross validation) ・Permutation test ・Bootstrapping (b) Visualization of cross-validation results ・ROC curve ・Calibration plot ・Decile plot ・Calibration belt 21. The method for automating exploratory data analysis based on variable attributes according to claim 20, comprising:

25. [Analysis when all variables are repeated measurements] (G) when all the variables are repeatedly measured, the method includes the steps of: classifying the variables into a case where there are two repeatedly measured variables and a case where there are three or more repeatedly measured variables; classifying the variables into a case where the variables are continuous variables and a case where the variables are categorical variables; and when there are two repeatedly measured variables and the variables are continuous variables, the method includes the steps of performing a normal distribution analysis on the continuous variables, determining whether the continuous variables follow a normal distribution, and if they do follow a normal distribution, setting a paired sample T test, and if they do not follow a normal distribution, setting a Wilcoxon signed-rank test to perform the data analysis; When there are two repeated measurement variables and the repeated measurement variables are categorical variables, the method includes the steps of: performing data analysis by setting McNemar's test and Cohens' Kappa when the number of subgroups (m, n) included in the categorical variables is two (m=2, n=2); performing data analysis by setting Cochran-Armitage test for trend when the number of subgroups (m, n) included in the categorical variables is two, and three or more (m=2, n≧3); and performing data analysis by setting McNemar-Bowker test when the number of subgroups (m, n) included in the categorical variables is three or more (m≧3, n≧3), When the number of repeated measurement variables is three or more and the repeated measurement variables are continuous variables, the method includes a step of setting up a linear mixed effect model analysis, a GEE (Generalized Estimating Equation) analysis, or a Repeated Measures 1-way ANOVA to perform data analysis; 18. The method for automating exploratory data analysis based on variable attributes according to claim 16 or 17, further comprising the step of setting up Generalized mixed effect model analysis (GEE (Generalized Estimating Equation) analysis) to perform data analysis when there are three or more repeated measurement variables and the repeated measurement variables are categorical variables.

26. [Analysis when single measurement variables are mixed] (H) When a single measurement variable and a repeatedly measured variable are mixed, the repeatedly measured variables are classified into those in which the repeatedly measured variables are continuous variables and those in which the repeatedly measured variables are categorical variables. When the repeated measurement variables are continuous variables, a linear mixed effect model analysis or a GEE (Generalized Estimating Equation) analysis is set up to perform data analysis, and when the repeated measurement variables are continuous variables and the single measurement variable is a categorical variable, a Repeated Measures 2-way ANOVA is set up to perform data analysis; 18. The method for automating exploratory data analysis based on variable attributes according to claim 16 or 17, further comprising the step of setting up Generalized mixed effect model analysis (GEE (Generalized Estimating Equation) analysis) to perform data analysis when the repeated measurement variables are categorical variables.

27. [Analysis of Survival Materials] (I) Analysis of survival data is divided into cases where only survival time and event occurrence variables are available and cases where survival time, event occurrence variables, and single measurement data are available.

18. The method for automating exploratory data analysis based on variable attributes according to claim 16, comprising the steps of: setting a Kaplan-Meier curve analysis algorithm to perform data analysis when only survival time and event occurrence variables are available; and setting a Univariable Cox proportional hazards regression to analyze the individual influence of a single measurement variable on survival when survival time, event occurrence data, and single measurement data are available; setting a Multivariable Cox proportional hazards regression to analyze the integrated influence of a single measurement variable on survival; and setting a Kaplan-Meier curve analysis to perform Log rank (comparison of survival probability differences between detailed categories).

28. The method for automating exploratory data analysis based on variable attributes according to claim 27, further comprising: a step of constructing a survival probability prediction model when survival time, event occurrence variables, and single measurement data are available; and a step of analyzing the survival probability prediction model by performing a discrimination aspect prediction performance analysis at time=t, a calibration aspect prediction performance analysis at time=t, and a survival probability prediction model cross-validation analysis.

Images