Elderly health data knowledge analysis method and system

Abstract

The present invention provides a method and system for analyzing knowledge of elderly health data, and relates to the technical field of data processing. The present invention proposes a feature selection method based on ensemble learning, constructs a knowledge discovery model, and simultaneously integrates machine learning evaluation indicators and interpretive machine learning evaluation methods, so that the knowledge discovery model of elderly health survey data has both evaluation indicators and interpretive evaluation methods, and can analyze the relationship between independent variables and dependent variables in elderly health survey data from multiple dimensions, so as to accurately and comprehensively analyze the factors affecting the health of the elderly.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for analyzing knowledge of elderly health data. Background Art

[0002] Elderly health data include personal data on daily activities, instrumental activities, education, work type, exercise habits, dietary health, medical services, psychological status, etc. Elderly health data are complex, accurate, safe, heterogeneous and closed. Comprehensive analysis and research on factors affecting the health of the elderly is of great significance for scientifically understanding the health problems of aging and achieving healthy aging.

[0003] For a long time, the analysis method of health census data has been statistical methods or simple linear regression analysis, that is, proposing hypotheses and then mining and verifying the relationships in the data.

[0004] However, although the existing analysis methods of health census data based on statistical methods or simple linear regression analysis can make full use of limited valuable data, the assumptions put forward are generally based on a certain major factor, such as residential pattern, education level, economic status, etc., and are not considered comprehensively in combination with specific factors in multiple aspects. There are certain limitations, resulting in low accuracy of the analysis results. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies in the prior art, the present invention provides a method and system for analyzing knowledge of elderly health data, which solves the technical problem of low accuracy of analysis results.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] The present invention provides a method for analyzing knowledge of elderly health data, which is executed by a computer and comprises the following steps:

[0010] S1. Obtain elderly health survey data;

[0011] S2. Preprocessing the elderly health survey data;

[0012] S3, perform integrated feature selection on the pre-processed elderly health survey data to obtain key features;

[0013] S4, acquiring a knowledge discovery model based on the key features;

[0014] S5. Based on the machine learning evaluation index, the explanation machine learning evaluation method and the knowledge discovery model, a knowledge discovery model for elderly health survey data is obtained;

[0015] S6. Analyze the elderly health survey data based on the elderly health survey data knowledge discovery model to obtain the relationship between the independent variable and the dependent variable in the elderly health survey data.

[0016] Preferably, the preprocessing includes missing value processing.

[0017] Preferably, the missing value processing includes:

[0018] S201, let a set of elderly health survey data be x 1 ......x k , which represents the missing values ​​in this set of elderly health survey data in a unified form;

[0019] S202, starting from the first elderly health survey data, if x 1 There are missing values ​​in other variables x 2 to x k On the regression, x 1 The missing values ​​in are replaced by the predicted values ​​obtained from x 1 The posterior predictive distribution of is simulated;

[0020] S203, for the next elderly health survey data x with missing values i , in all other elderly health survey data x 1 ......x k Upper regression, x i The missing values ​​in x are replaced by i All missing values ​​in the elderly health survey data are replaced by simulated values ​​of the posterior predictive distribution.

[0021] Preferably, the integrated feature selection includes:

[0022] S301, respectively using chi-square test, mutual information, L2-based logistic regression, random logistic regression, random forest and Light GBM to perform feature selection on a set of elderly health survey data after preprocessing, and obtain 6 types of feature score vectors of feature a;

[0023] S302: normalize the 6 types of feature score vectors of the feature a. The normalization formula includes:

[0024]

[0025] in:

[0026] X * is the normalized feature score vector;

[0027] X i is one of the six types of feature score vectors of feature a;

[0028] S303, obtaining a feature average total score value based on the 6 types of feature score vectors of the normalized feature a;

[0029] S304. Based on the method of steps S301 to S303, obtain the feature average total score of M features in the elderly health survey data, then add the feature average total score of the M features to obtain a total score, divide the total score by 6, and then divide it by the number of elderly people in the elderly health survey data to obtain a final average total score, and select features in the elderly health survey data whose feature average total score is above the final average total score as key features.

[0030] Preferably, the method for acquiring the knowledge discovery model includes:

[0031] The key features are input into the gradient boosted tree classifier to obtain the knowledge discovery model.

[0032] Preferably, the specific method for obtaining the knowledge discovery model of elderly health survey data includes:

[0033] S501, evaluating the knowledge discovery model based on the machine learning evaluation index, and screening the knowledge discovery model based on the evaluation result to obtain the initial knowledge discovery model of the elderly health survey data;

[0034] S502. Two explanation machine learning evaluation methods, namely, ranking importance and Shapley Additive Explanations based on game theory Shapley value, are added to the initial knowledge discovery model of the elderly health survey data to obtain the knowledge discovery model of the elderly health survey data.

[0035] Preferably, in step S501, the machine learning evaluation indicators include: accuracy, F value and AUC value.

[0036] The present invention also provides a system for analyzing knowledge of elderly health data, the system comprising a computer, the computer comprising:

[0037] at least one storage unit;

[0038] at least one processing unit;

[0039] The at least one storage unit stores at least one instruction, and the at least one instruction is loaded and executed by the at least one processing unit to implement the following steps:

[0040] S1. Obtain elderly health survey data;

[0041] S2. Preprocessing the elderly health survey data;

[0042] S3, perform integrated feature selection on the pre-processed elderly health survey data to obtain key features;

[0043] S4, acquiring a knowledge discovery model based on the key features;

[0044] S5. Based on the machine learning evaluation index, the explanation machine learning evaluation method and the knowledge discovery model, a knowledge discovery model for elderly health survey data is obtained;

[0045] S6. Analyze the elderly health survey data based on the elderly health survey data knowledge discovery model to obtain the relationship between the independent variable and the dependent variable in the elderly health survey data.

[0046] (III) Beneficial effects

[0047] The present invention provides a method and system for analyzing knowledge of elderly health data. Compared with the prior art, it has the following beneficial effects:

[0048] The present invention performs integrated feature selection on the features in the elderly health survey data after preprocessing to obtain key features, and then obtains a knowledge discovery model based on the key features, and integrates the machine learning evaluation index and the explanatory machine learning evaluation method into the knowledge discovery model to obtain the elderly health survey data knowledge discovery model, and then analyzes the relationship between the independent variable and the dependent variable in the elderly health survey data through the elderly health survey data knowledge discovery model. The present invention proposes a feature selection method based on integrated learning, constructs a knowledge discovery model, and integrates the evaluation method based on the evaluation index and the explanatory machine learning evaluation method, so that the elderly health survey data knowledge discovery model has both evaluation indicators and explanatory evaluation methods, and can analyze the relationship between the independent variable and the dependent variable in the elderly health survey data from multiple dimensions, so as to accurately and comprehensively analyze the factors affecting the health of the elderly. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A block diagram of a method for analyzing knowledge of elderly health data according to an example of the present invention;

[0051] Figure 2 It is a ROC curve diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The embodiments of the present application solve the problem of low accuracy of the analysis methods of health census data in the prior art by providing a method and system for analyzing knowledge of elderly health data, and analyze the relationship between independent variables and dependent variables in elderly health survey data from multiple dimensions, thereby accurately and comprehensively analyzing the factors affecting the health of the elderly.

[0054] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0055] The embodiment of the present invention performs integrated feature selection on the features in the elderly health survey data after preprocessing to obtain key features, then obtains a knowledge discovery model based on the key features, and integrates the evaluation method based on evaluation indicators and explanatory machine learning into the knowledge discovery model to obtain a knowledge discovery model for elderly health survey data, and then analyzes the relationship between independent variables and dependent variables in the elderly health survey data through the knowledge discovery model for elderly health survey data. The embodiment of the present invention proposes a feature selection method based on integrated learning, constructs a knowledge discovery model, and integrates the evaluation method based on evaluation indicators and explanatory machine learning, so that the knowledge discovery model for elderly health survey data has both machine learning evaluation indicators and explanatory evaluation methods, and can analyze the relationship between independent variables and dependent variables in elderly health survey data from multiple dimensions, so as to accurately and comprehensively analyze the factors affecting the health of the elderly.

[0056] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0057] The embodiment of the present invention provides a method for analyzing knowledge of elderly health data, which is executed by a computer. Figure 1 As shown, steps S1 to S6 are included:

[0058] S1. Obtain elderly health survey data;

[0059] S2. Preprocessing of elderly health survey data;

[0060] S3, perform integrated feature selection on the pre-processed elderly health survey data to obtain key features;

[0061] S4, obtaining a knowledge discovery model based on the above key features;

[0062] S5. Based on the machine learning evaluation index, the explanation machine learning evaluation method and the knowledge discovery model, a knowledge discovery model for elderly health survey data is obtained;

[0063] S6. Analyze the elderly health survey data based on the elderly health survey data knowledge discovery model to obtain the relationship between the independent variables and the dependent variables in the elderly health survey data.

[0064] The embodiment of the present invention proposes a feature selection method based on ensemble learning, constructs a knowledge discovery model, and integrates machine learning evaluation indicators and explanatory machine learning evaluation methods, so that the knowledge discovery model of the elderly health survey data has both machine learning evaluation indicators and explanatory evaluation methods, and can analyze the relationship between independent variables and dependent variables in the elderly health survey data from multiple dimensions, so as to accurately and comprehensively analyze the factors affecting the health of the elderly.

[0065] The following is a detailed description of each step:

[0066] S1. Obtain elderly health survey data. It should be noted that, in the embodiment of the present invention, the elderly health survey data takes the latest survey data of the CLHLS project in 2014 as an example. The original sample of the data contains 7192 samples, totaling 886 variables. Through the instruction manual of the survey, it is found that some variables are designed to ensure the rationality of the questionnaire survey, but these variables themselves have no practical meaning. For example, before each type of question, there will be a question about whether the question is answered by the person himself or by someone else. Each such question corresponds to a variable. Usually, the name code of such variables starts with the letter "r". We first delete such variables that have no practical significance for the research of this article, and then delete all samples with empty rows or columns.

[0067] S2. Preprocess the elderly health survey data. The specific implementation process is as follows:

[0068] S201, let a set of elderly health survey data be x 1 ......x k , which represents the missing values ​​in this set of elderly health survey data in a unified form;

[0069] S202, starting from the first elderly health survey data, if x 1 There are missing values, which will be in other variables x 2 to xk On the regression, x 1 The missing values ​​in are replaced by the predicted values ​​obtained from x 1 The posterior predictive distribution of is simulated;

[0070] S203, for the next elderly health survey data x with missing values i , in all other elderly health survey data x 1 ......x k Upper regression, x i The missing values ​​in x are replaced by i All missing values ​​in the elderly health survey data are replaced by simulated values ​​of the posterior predictive distribution.

[0071] It should be noted that, in the embodiment of the present invention, the variables with missing values ​​accounting for more than 10% are deleted. It should also be noted that the samples with an age of more than 105 are excluded in the embodiment of the present invention. The self-assessed health data set thus obtained is 6312, with 206 features (variables), and the mental health data set is 6314, with 206 features (variables). Then, feature selection is performed on the two data sets respectively to facilitate modeling. The following is a further analysis taking the self-assessed health data set as an example.

[0072] S3. Perform integrated feature selection on the pre-processed elderly health survey data to obtain key features. The specific implementation process is as follows:

[0073] S301, respectively using chi-square test, mutual information, L2-based logistic regression, random logistic regression, random forest and Light GBM to select a feature (in this embodiment of the present invention, a feature refers to feature a, feature a refers to one of M features, M=206) in a set of self-assessed health data sets after preprocessing, and obtain 6 types of feature score vectors for one feature;

[0074] S302, normalize the 6 types of feature score vectors. In order to ensure that the feature score vectors generated by different feature selection methods are comparable, it is necessary to perform normalization before performing feature score vector aggregation and fusion. The feature score vectors generated by each feature selection method are standardized to a range of [0, 1]. The normalization formula includes:

[0075]

[0076] in:

[0077] X * is the normalized feature score vector;

[0078] Xi It is one of the six types of feature score vectors;

[0079] S303: Obtain a feature average total score value based on the normalized 6 types of feature score vectors of a feature. Specifically, add the 6 types of feature score vectors of a feature and divide by 6 to obtain the feature average total score value.

[0080] S304. Based on the method of steps S301 to S303, the feature average total score of 206 features in the elderly health survey data is obtained, and then the feature average total score of the 206 features is added to obtain the total score, and the total score is divided by 6, and then divided by 6312 of the elderly in the elderly health survey data to obtain the final average total score, and the features with the feature average total score in the elderly health survey data above the final average total score are selected as key features. Finally, 44 key features are selected. As shown in Table 1. Among them, mean in the table refers to the final average total score, Stability refers to random logistic regression, Chi2 refers to chi-square test, Rf refers to random forest, lasso_l2 refers to L2-based logistic regression, Lgbm refers to Light GBM, and MI refers to mutual information.

[0081] Table 1

[0082]

[0083]

[0084] From the table, we can find that in terms of the results of feature selection, the chi-square test method has a large number of zero scores, while other methods generally maintain a relatively consistent trend, thereby screening out the most effective key features. Since the variables of the original sample are encoded in the form of letter codes and the meaning is unclear, the selected features (variables) are renamed according to their actual meanings, and the specific coding meaning of each feature (variable) is presented, as shown in Table 2:

[0085] Table 2

[0086]

[0087]

[0088]

[0089] Combined with the actual meaning of each feature, it can be seen that among the factors that determine the physical health of the elderly, the psychological state occupies a very important part. Several variables are related to the psychological state, such as whether they are still optimistic, anxious, lonely, etc. In addition, several chronic diseases that most often plague the elderly can be found, such as lung-related diseases, cardiovascular diseases, diabetes and joint diseases. These high-incidence elderly diseases often have a practical impact on the lives of the elderly. Some dietary habits also seem to secretly affect the physical health of the elderly, such as egg intake and milk intake. The above feature selection results only preliminarily indicate the possible importance of the features. The final ranking must be given after the model is trained.

[0090] S4. Obtain a knowledge discovery model based on the above key features. Specifically, the 44 key features are input into a gradient boosting tree classifier as input variables, and self-assessed health is used as a dependent variable to obtain a knowledge discovery model. At the same time, in the embodiment of the present invention, the knowledge discovery model is also obtained through commonly used models with high transparency, namely, logistic regression model, decision tree model, and random forest model.

[0091] S5. Based on the evaluation indicators, the interpretation of machine learning evaluation methods and the above-mentioned knowledge discovery model, obtain the knowledge discovery model of the elderly health survey data. The specific implementation process is as follows:

[0092] S501, evaluate the performance of the knowledge discovery model using machine learning evaluation indicators, where the machine learning evaluation indicators include accuracy, F value and AUC value. The knowledge discovery model is selected to obtain an initial knowledge discovery model for the elderly health survey data.

[0093] From Table 3 and Figure 2 It can be concluded that Table 3 and Figure 2 GBDT refers to the knowledge discovery model built by gradient boosting tree (GBDT). Combining the results of various models, it can be seen that the knowledge discovery model built by gradient boosting tree has the best performance in various indicators, especially in the key indicators of accuracy, F value and Auc (area under the ROC curve), and its comprehensive performance is better than other models. This shows that the knowledge discovery model built by gradient boosting tree can identify the self-assessed health status of the elderly with good efficiency, whether it is healthy or unhealthy.

[0094] Table 3

[0095] Accuracy Precision Recall F1 Auc Logistic Regression 76.22％ 73.11％ 72.89％ 72.97％ 83.79％ Decision Tree 74.72％ 69.48％ 75.76％ 72.48％ 81.78％ Random Forest 74.78％ 75.17％ 63.90％ 69.04％ 81.65％ GBDT 77.74％ 74.48％ 75.37％ 74.89％ 85.19％

[0096] S502. Two explanation machine learning evaluation methods, namely, permutation importance and Shapley Additive Explanations based on game theory Shapley value, are added to the initial knowledge discovery model of the elderly health survey data to obtain the knowledge discovery model of the elderly health survey data.

[0097] S6. Analyze the above elderly health survey data based on the above elderly health survey data knowledge discovery model to obtain the relationship between the independent variables and the dependent variables in the above elderly health survey data. The specific implementation process is as follows:

[0098] First, we examine how each factor affects the self-assessment of health of the elderly individually. The results obtained by the permutation importance method are shown in Table 4. In the table, the closer the feature is to the top, the greater the impact on the prediction result, and the feature at the bottom has the least impact. Because the calculation is performed by random shuffling, multiple operations are actually performed. The impact size is expressed by the mean plus the variance. If a value appears to be negative, it means that the prediction result of the feature after random shuffling is better. Such features are basically useless and can be considered to be deleted afterwards.

[0099] Observing the top ten rankings, in the embodiment of the present invention, it can be seen that the most important thing is the evaluation of life satisfaction, which is also a characterization variable with a high correlation with mental health. This shows that the health of the elderly must be comprehensive health, and both psychological and physiological aspects are indispensable. At the same time, the elderly's feelings about the changes in their health status are also very important. This requires attention to communication with the elderly and their expression of their feelings about their physical condition, which will be a very important health signal. At the same time, sleep quality and mentality, especially a positive and optimistic mentality, also occupy a relatively important position. Limited activity and emergency medical treatment are often direct manifestations of problems in the health status of the elderly, and whether they can squat and stand up is a manifestation of the weakness of the elderly, so they also have a higher ranking. The age of the elderly also plays a very important role, indicating that health status may be related to the distribution of age, and it is necessary to provide corresponding health services for different age groups in a targeted manner. In the top 20, it is also noted that the height of the elderly also appears in a relatively forward position. Through the search of domestic and foreign literature on the relationship between the height and health of the elderly, it is found that some foreign studies point out that as the elderly age, their height will shrink to a certain extent, which has a very important relationship with the physiological health of the elderly and is an important signal of the physiological health status of the elderly. This is something that previous researchers have not covered in their research on the survey data set. In particular, this provides a possible simple way to continuously track the health of the elderly, that is, by paying attention to the changes in the height of the elderly, as one of the important signals for warning of the health of the elderly. Lung diseases such as heart disease and asthma are the most prominent chronic diseases of the elderly. The number of teeth, the ability to walk a thousand meters and bend down to pick up things are also important manifestations of the physical health of the elderly, and the importance ranking in the embodiment of the present invention is relatively high. In addition, the province is also in the top 20, indicating that the self-assessment of health in the elderly may also have a strong relationship with the geographical distribution. In the final ranking, two negative values ​​appeared, which means that the two variables have no effect on predicting self-assessed health, one of which is family income, which seems to be inconsistent with the intuition of most people, because most people generally believe that family income is higher, there will be better medical care, and health should also be better.

[0100] Table 4

[0101]

[0102]

[0103] The permutation importance explanation method has a fast calculation speed and can quickly give a global importance ranking of the feature, reflecting whether the feature is important to the model. However, this influence is not intuitive enough. For example, when the permutation importance value is medium, it may mean that it only has a large impact on a small number of predictions, but the overall impact is small, or it has a medium impact on all predictions, that is, the permutation importance cannot reflect how each feature specifically affects the final result. Therefore, in the example of the present invention, the Shapley Additive Explanations method based on the game theory Shapley value is also introduced. The calculation of this method is more complicated, but this method can specifically measure the role of each feature in each sample prediction or all sample predictions, and provide local insights into the importance of features. At the same time, these two explanatory methods can confirm each other, which helps to enhance trust in the model and results.

[0104] The embodiment of the present invention further provides a system for analyzing knowledge of elderly health data, the system comprising a computer, the computer comprising:

[0105] at least one storage unit;

[0106] at least one processing unit;

[0107] The at least one storage unit stores at least one instruction, and the at least one instruction is loaded and executed by the at least one processing unit to implement the following steps:

[0108] S1. Obtain elderly health survey data;

[0109] S2. Preprocessing of elderly health survey data;

[0110] S3, perform integrated feature selection on the pre-processed elderly health survey data to obtain key features;

[0111] S4, acquiring a knowledge discovery model based on the key features;

[0112] S5. Based on the machine learning evaluation index, the explanation machine learning evaluation method and the above-mentioned knowledge discovery model, obtain the knowledge discovery model of the elderly health survey data;

[0113] S6. Analyze the above elderly health survey data based on the above elderly health survey data knowledge discovery model to obtain the relationship between the independent variables and the dependent variables in the above elderly health survey data.

[0114] It can be understood that the above-mentioned elderly health data knowledge analysis system provided by the embodiment of the present invention corresponds to the above-mentioned elderly health data knowledge analysis method. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the elderly health data knowledge analysis method, and will not be repeated here.

[0115] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0116] The embodiment of the present invention proposes an integrated feature selection method to construct a knowledge discovery model, and at the same time integrates machine learning evaluation indicators and explanatory machine learning evaluation methods, so that the knowledge discovery model of the elderly health survey data has both machine learning evaluation indicators and explanatory evaluation methods, and can analyze the relationship between independent variables and dependent variables in the elderly health survey data from multiple dimensions, so as to accurately and comprehensively analyze the factors affecting the health of the elderly.

[0117] It should be noted that, through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0119] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing knowledge of elderly health data, characterized in that: The method is executed by a computer and comprises the following steps: S1. Obtaining elderly health survey data, including the survey data of the CLHLS project in 2014; S2. Preprocessing the elderly health survey data; S3, performing feature selection based on ensemble learning on the preprocessed elderly health survey data to obtain key features; S4, obtaining a knowledge discovery model based on the key features, including inputting the key features into a gradient boosting tree classifier to obtain a knowledge discovery model; S5. Based on the machine learning evaluation index, the explanation machine learning evaluation method and the knowledge discovery model, a knowledge discovery model of the elderly health survey data is obtained; wherein the specific method of obtaining the knowledge discovery model includes: S501, evaluating the knowledge discovery model based on the machine learning evaluation index, and screening the knowledge discovery model based on the evaluation result to obtain the initial knowledge discovery model of the elderly health survey data; S502, adding two explanation machine learning evaluation methods, namely, ranking importance and Shapley Additive Explanations based on game theory Shapley value, to the initial knowledge discovery model of elderly health survey data to obtain the knowledge discovery model of elderly health survey data; S6. Analyze the elderly health survey data based on the elderly health survey data knowledge discovery model to obtain the relationship between the independent variable and the dependent variable in the elderly health survey data.

2. The elderly health data knowledge analysis method according to claim 1, characterized in that: The preprocessing includes missing value processing.

3. The elderly health data knowledge analysis method according to claim 2, characterized in that: The missing value processing includes: S201, let a set of elderly health survey data be x1......x k , represents the missing values ​​in this set of elderly health survey data in a unified form; S202, starting from the first elderly health survey data, if x1 has missing values, in other variables x2 to x k Upper regression, replacing missing values ​​in x1 with predicted values ​​simulated from the posterior predictive distribution of x1; S203, for the next elderly health survey data x with missing values i , in all other elderly health survey data x1......x k Upper regression, x i The missing values ​​in x are replaced by i All missing values ​​in the elderly health survey data are replaced by simulated values ​​of the posterior predictive distribution.

4. The elderly health data knowledge analysis method according to claim 1, characterized in that: The feature selection based on ensemble learning includes: S301, respectively using chi-square test, mutual information, L2-based logistic regression, random logistic regression, random forest and LightGBM to perform feature selection on a set of elderly health survey data after preprocessing, and obtain 6 types of feature score vectors of feature a; S302: normalize the 6 types of feature score vectors of the feature a. The normalization formula includes: in: X * is the normalized feature score vector; X i is one of the six types of feature score vectors of feature a; S303, obtaining a feature average total score value based on the 6 types of feature score vectors of the normalized feature a; S304. Based on the method of steps S301 to S303, obtain the feature average total score of M features in the elderly health survey data, then add the feature average total score of the M features to obtain a total score, divide the total score by 6, and then divide it by the number of elderly people in the elderly health survey data to obtain a final average total score, and select features in the elderly health survey data whose feature average total score is above the final average total score as key features.

5. The elderly health data knowledge analysis method according to claim 1, characterized in that: In step S501, the machine learning evaluation indicators include: accuracy, F value and AUC value.

6. An elderly health data knowledge analysis system, characterized in that: The system includes a computer, wherein the computer includes: at least one storage unit; at least one processing unit; The at least one storage unit stores at least one instruction, and the at least one instruction is loaded and executed by the at least one processing unit to implement the following steps: S1. Obtaining elderly health survey data, including the survey data of the CLHLS project in 2014; S2. Preprocessing the elderly health survey data; S3, performing feature selection based on ensemble learning on the preprocessed elderly health survey data to obtain key features; S4, obtaining a knowledge discovery model based on the key features, including inputting the key features into a gradient boosting tree classifier to obtain a knowledge discovery model; S5. Based on the machine learning evaluation index, the explanation machine learning evaluation method and the knowledge discovery model, a knowledge discovery model of the elderly health survey data is obtained; wherein the specific method of obtaining the knowledge discovery model includes: S501, evaluating the knowledge discovery model based on the machine learning evaluation index, and screening the knowledge discovery model based on the evaluation result to obtain the initial knowledge discovery model of the elderly health survey data; S502, adding two explanation machine learning evaluation methods, namely, ranking importance and Shapley Additive Explanations based on game theory Shapley value, to the initial knowledge discovery model of elderly health survey data to obtain the knowledge discovery model of elderly health survey data; S6. Analyze the elderly health survey data based on the elderly health survey data knowledge discovery model to obtain the relationship between the independent variable and the dependent variable in the elderly health survey data.

Images