Feature screening method and device, storage medium and electronic equipment

Abstract

The embodiment of the invention provides a feature screening method and device, a storage medium and electronic equipment, and the method comprises the steps: obtaining an initial feature set and a target dependent variable; taking each abnormal value threshold in the plurality of abnormal value thresholds as a current abnormal value threshold, and executing the following feature screening operation to obtain a plurality of feature sets: according to a correlation coefficient between each initial independent variable and a target dependent variable and the current abnormal value threshold, performing linear regression processing on each initial independent variable in sequence to obtain a plurality of feature sets; and obtaining a saliency probability value of each initial independent variable under the current abnormal value threshold, so as to screen the initial independent variables from the initial feature set, obtain a feature set corresponding to the current abnormal value threshold, and select a target feature set. The problem that the reliability of a prediction model is low when data containing abnormal values are used for training is solved, and the effect of improving the prediction efficiency of the model is achieved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of data mining. Specifically, the embodiments of the present application relate to a feature screening method, apparatus, storage medium, and electronic device. Background Art

[0002] In the fields of data mining and machine learning, feature screening is one of the key steps in constructing a prediction model, aiming to identify the set of independent variables that have the most influence on predicting the target dependent variable. However, actual datasets are often contaminated by outliers or noise. The existence of these outliers will seriously distort the results of the least squares estimation and other regression models based on similar assumptions, resulting in the distortion of model parameters, thereby affecting the significance judgment of features and the prediction performance of the model.

[0003] In related technologies, a common practice for dealing with outliers is to identify and remove outliers before model construction, such as using descriptive statistics, the isolation forest algorithm, or deep learning-based methods. However, these methods not only increase the complexity of data preprocessing but also may bring additional problems, such as hypothesis bias, overfitting risk, and reduced model interpretability, which lead to the problem that the reliability of the prediction model is relatively low when using data containing outliers for training in related technologies. Summary of the Invention

[0004] The embodiments of the present application provide a feature screening method, apparatus, storage medium, and electronic device to at least solve the technical problem that the reliability of the prediction model is relatively low when using data containing outliers for training in related technologies.

[0005] According to one aspect of the embodiments of the present application, a feature screening method is provided, including:

[0006] Obtain an initial feature set and a target dependent variable, where each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable; determine a plurality of outlier thresholds according to an outlier threshold range and a threshold adjustment step size, and use each outlier threshold in the plurality of outlier thresholds as the current outlier threshold, and perform the following feature screening operations to obtain a plurality of feature sets, where each feature set in the plurality of feature sets corresponds to one outlier threshold in the plurality of outlier thresholds respectively: perform linear regression processing on each initial independent variable in turn according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, to obtain the significance probability value of each initial independent variable under the current outlier threshold; screen the initial independent variables from the initial feature set according to the significance probability value of each initial independent variable under the current outlier threshold, to obtain the feature set corresponding to the current outlier threshold; select a target feature set from the plurality of feature sets, where the target feature set is used to train a prediction model corresponding to the target dependent variable.

[0007] According to another aspect of the embodiments of the present application, there is also provided a feature screening device, including:

[0008] An acquisition unit, configured to acquire an initial feature set and a target dependent variable, where each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable;

[0009] A screening unit, configured to determine a plurality of outlier thresholds according to an outlier threshold range and a threshold adjustment step size, and use each outlier threshold in the plurality of outlier thresholds as the current outlier threshold, and perform the following feature screening operations to obtain a plurality of feature sets, where each feature set in the plurality of feature sets corresponds to one outlier threshold in the plurality of outlier thresholds respectively: perform linear regression processing on each initial independent variable in turn according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, to obtain the significance probability value of each initial independent variable under the current outlier threshold; screen the initial independent variables from the initial feature set according to the significance probability value of each initial independent variable under the current outlier threshold, to obtain the feature set corresponding to the current outlier threshold;

[0010] A selection unit, configured to select a target feature set from the plurality of feature sets, where the target feature set is used to train a prediction model corresponding to the target dependent variable.

[0011] According to another aspect of the embodiments of the present application, there is also provided a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any one of the above method embodiments when running.

[0012] According to another aspect of the embodiments of the present application, there is provided a computer program product or a computer program. The computer program product or the computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the steps in any one of the above method embodiments.

[0013] According to another aspect of the embodiments of the present application, there is also provided an electronic device including a memory and a processor. A computer program is stored in the memory, and the processor is configured to execute the steps in any one of the above method embodiments through the computer program.

[0014] Through the present application, according to the outlier threshold range and the threshold adjustment step size, a plurality of outlier thresholds are determined, and each outlier threshold in the plurality of outlier thresholds is used as the current outlier threshold to perform a feature screening operation, obtaining a plurality of feature sets, where each feature set in the plurality of feature sets respectively corresponds to one outlier threshold in the plurality of outlier thresholds: according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, each initial independent variable is sequentially subjected to linear regression processing to obtain the significance probability value of each initial independent variable under the current outlier threshold; according to the significance probability value of each initial independent variable under the current outlier threshold, initial independent variables are screened from the initial feature set to obtain the feature set corresponding to the current outlier threshold, so as to select a target feature set from the plurality of feature sets by combining the correlation coefficient and the significance probability value to train a prediction model, which can more accurately identify those independent variables that not only have a strong numerical correlation but also make a significant contribution to the prediction model in a statistical sense, so as to reduce the influence of outliers on the training of the prediction model, solve the problem that the reliability of the prediction model obtained by training using these data is relatively low in the related art, and through the iterative screening process, can more accurately identify the feature set that has a significant predictive effect on the dependent variable and eliminate non-significant features at the same time, thereby improving the interpretability and prediction efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a schematic diagram of an application scenario of a feature screening method according to an embodiment of the present application;

[0016] Figure 2It is a schematic flowchart of an optional feature screening method according to an embodiment of the present application;

[0017] Figure 3 It is a schematic flowchart of another optional feature screening method according to an embodiment of the present application;

[0018] Figure 4 It is a structural block diagram of an optional feature screening device according to an embodiment of the present application;

[0019] Figure 5 It is a computer system structural block diagram of an optional electronic device according to an embodiment of the present application. Detailed implementation manners

[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order different from those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0022] According to one aspect of the embodiments of the present application, a feature screening method is provided. Optionally, in this embodiment, the above-mentioned feature screening method can be but is not limited to being applied to a hardware environment including a terminal device 102 and a server 104 as Figure 1 shown. The server 104 can be connected to the terminal device 102 through a network and can be used to provide services (such as application services, etc.) for the terminal device 102 or a client installed on the terminal device 102. A database can be set on the server 104 or independently of the server 104 to provide data storage services for the server 104.

[0023] The above network may include, but is not limited to, at least one of the following: a wired network, a wireless network. The above wired network may include, but is not limited to, at least one of the following: a wide area network, a metropolitan area network, a local area network. The above wireless network may include, but is not limited to, at least one of the following: WIFI (Wireless Fidelity), Bluetooth. The terminal device 102 may be, but is not limited to, a PC (Personal Computer), a mobile phone, a tablet computer, etc. The server 104 may be, but is not limited to, a cloud server, a server cluster, or other server types.

[0024] The feature screening method of the embodiments of the present application may be executed by the server 104, or may be executed by the terminal device 102, or may also be jointly executed by the server 104 and the terminal device 102. Among them, when the terminal device 102 executes the feature screening method of the embodiments of the present application, it may also be executed by the client installed thereon.

[0025] Taking the execution of the feature screening method in this embodiment by the terminal device 102 as an example, Figure 2 is a schematic flowchart of an optional feature screening method according to an embodiment of the present application, as Figure 2 shown, the process of this method may include the following steps:

[0026] Step S202, obtain an initial feature set and a target dependent variable. Among them, each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable;

[0027] Step S204, according to the outlier threshold range and the threshold adjustment step size, determine multiple outlier thresholds, and use each outlier threshold in the multiple outlier thresholds as the current outlier threshold, and perform the following feature screening operations to obtain multiple feature sets. Among them, each feature set in the multiple feature sets respectively corresponds to one outlier threshold in the multiple outlier thresholds: according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, perform linear regression processing on each initial independent variable in turn to obtain the significance probability value of each initial independent variable under the current outlier threshold; according to the significance probability value of each initial independent variable under the current outlier threshold, screen the initial independent variables from the initial feature set to obtain the feature set corresponding to the current outlier threshold;

[0028] Step S206, select a target feature set from the multiple feature sets. Among them, the target feature set is used to train a prediction model corresponding to the target dependent variable.

[0029] The feature screening method in this embodiment can be applied to the fields of data mining technology and machine learning, and is applied to scenarios where datasets are easily interfered by outliers, such as financial risk assessment, credit scoring, medical data analysis, Internet of Things monitoring, industrial quality control, etc. For example, in the field of financial risk assessment, when banks and financial institutions evaluate the credit status of loan applicants, they will collect a large amount of information about the applicants, including income, liabilities, age, occupation, etc. These data may contain outliers for various reasons, such as data entry errors, system failures, or extreme situations, and these outliers will have a negative impact on the accuracy of the credit scoring model.

[0030] In related technologies, a common practice for dealing with outliers is to identify and remove outliers before model construction. For example, descriptive statistics, isolation forest algorithms, or deep learning-based methods are used. However, these methods not only increase the complexity of data preprocessing, but may also bring additional problems, such as hypothesis bias, overfitting risk, and reduction of model interpretability, which lead to the problem that the reliability of the prediction model obtained by training with these data in related technologies is relatively low.

[0031] To at least partially solve the above technical problems, in this embodiment, by dynamically adjusting the outlier threshold and combining the feature screening strategy of the significance probability value, the problem that the reliability of the prediction model is relatively low when using data containing outliers for training in related technologies is solved. It not only considers the correlation between the independent variable and the target dependent variable, but also combines the influence of the outlier threshold on the model performance. Through the iterative screening process, it can more accurately identify the feature set that has a significant predictive effect on the dependent variable, and at the same time eliminate non-significant features, thereby improving the interpretability and prediction efficiency of the model. Different from the traditional preprocessing step that requires manual identification and removal of outliers, the embodiment of this application can automatically adjust the sensitivity to outliers during the model training process, thus avoiding model bias and overfitting problems caused by improper outlier identification and processing.

[0032] It should be noted that the initial feature set can be a set containing all potential predictive variables. The initial feature set is usually composed of all independent variables that may affect the target dependent variable. For example, when evaluating the housing price, the initial feature set may include variables such as housing area, number of rooms, geographical location, age, and decoration status. Among them, each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable, and each initial independent variable includes multiple data belonging to the same type.

[0033] The target dependent variable can be the target demand variable to be predicted, and can include multiple data belonging to the same type. It is usually represented by (y). In the above example of housing price evaluation, the target dependent variable is the housing price.

[0034] The outlier threshold can be a predefined value used to determine whether a data point is an outlier. When the residual between the data point and the model predicted value exceeds this threshold, the point is considered an outlier. For example, if the outlier threshold is set to three standard deviations, then all data points with residuals greater than three standard deviations will be considered outliers.

[0035] The threshold adjustment step size can be the increment or decrement used when determining the outlier threshold. By setting different step sizes, feature screening operations can be iteratively performed among a series of thresholds to find the best outlier handling strategy. For example, the step size can be set to 0.5 and incremented step by step starting from 1 until the optimal outlier threshold is found.

[0036] The correlation coefficient can be used to evaluate the strength and direction of the linear relationship between two variables. Specifically, it evaluates the linear correlation strength and direction between the initial independent variable and the target dependent variable. Generally, the Pearson correlation coefficient can be used as the correlation coefficient.

[0037] The significance probability value is the probability value used to represent the result of a hypothesis test. The significance probability value is also called the P - value and is used to evaluate the significance of the relationship between the independent variable and the dependent variable. Usually, the P - value can evaluate the probability of observing the statistical result (such as the t - statistic) or a more extreme result if the independent variable and the dependent variable actually have no relationship. For example, a low P - value (such as less than 0.05) indicates that the relationship between the independent variable and the dependent variable is statistically significant and should be retained.

[0038] Linear regression processing is a statistical method that can be used to establish a mathematical model of the relationship between one or more independent variables and a dependent variable. Optionally, the linear regression processing may include one or more methods, such as least squares method, ridge regression, Lasso regression, Huber regression, and so on. Specifically, the least squares method is a common linear regression fitting method, and the goal is to minimize the sum of the squares of the distances from all data points to the regression line. Ridge regression is a regularization technique that can avoid overfitting of the model by adding a penalty term of the sum of the squares of the independent variables to the loss function, and at the same time has a certain tolerance for outliers. Lasso (Least Absolute Shrinkage and Selection Operator) regression is also a regularization regression. It controls the complexity of the model by adding a penalty term of the absolute value of the independent variables to the loss function, and at the same time can also perform feature selection, that is, it can shrink the weights of unimportant features to zero, thereby eliminating these features. Huber regression is a robust regression method that combines the advantages of the least squares method and the least absolute value error. It uses a mixed loss function. For data points with residuals less than a certain threshold, the square loss is used; for data points with residuals greater than or equal to the threshold, the linear loss is used. This method can reduce the impact of outliers on the model.

[0039] Optionally, the process of obtaining the initial feature set and the target dependent variable may be: in response to the obtained feature processing instruction, obtain the target dependent variable indicated by the feature processing instruction; according to the target dependent variable, obtain the data corresponding to the target dependent variable from multiple data sources connected to the terminal device as the initial independent variables to form the initial feature set. In specific practice, when the problem to be solved or the prediction target is determined according to the target dependent variable and the user or system indicated by the feature processing instruction. For example, in a credit risk analysis project, the user may indicate the target dependent variable as "loan default status" through a feature processing instruction (which may be selected through the user interface, input through script parameters, or system configuration files, etc.). Once the system receives this instruction, it will identify the target dependent variable and prepare for subsequent feature screening and model training. After determining the target dependent variable, it is necessary to collect the independent variable data that may affect this dependent variable from relevant data sources. This may involve querying from a database, calling an external API, reading a file, or capturing real-time data streams. The multiple data sources may include different types of databases, data warehouses, cloud storage services, Internet of Things device sensor data, etc.

[0040] Optionally, both the outlier threshold range and the threshold adjustment step size can be selected based on experience for the target dependent variable. Among them, the selection of the outlier threshold range is usually based on prior knowledge and preliminary analysis of the dataset. Specifically, the following strategies can be adopted: the strategy based on the data distribution characteristics of the target dependent variable and the initial independent variables, the strategy based on empirical values, and the strategy based on the preliminary model fitting without outlier processing. Among them, the strategy based on the data distribution characteristics of the target dependent variable and the initial independent variables can conduct descriptive statistical analysis on the target dependent variable and the initial independent variables, such as calculating the mean, standard deviation, quartiles, etc., to understand the data distribution characteristics. The strategy based on the preliminary model fitting without outlier processing can use the preliminary model without outlier processing, and determine the outlier threshold range through the residual distribution between the prediction and the actual value. Finding the boundary of the outlier points in the residual distribution can be used as the definition of the threshold range.

[0041] The threshold adjustment step size can be selected based on experience. For example, an equal-interval step size can be selected. Of course, in specific practice, the step size can also be adaptively adjusted according to the situation of feature screening, or the step size based on the change rate of the loss function, etc. Specifically, if the outlier distribution in the dataset is uneven, or the preliminary model fitting results show that the influence of outliers is more significant near certain thresholds, the adaptive adjustment step size strategy can be adopted. Of course, the step size can also be adjusted according to the rate of change of the loss function with the threshold. When the change rate of the loss function is high, it indicates that the threshold adjustment has a greater impact on the model performance. At this time, a smaller step size can be selected; on the contrary, when the change rate is low, the step size can be increased.

[0042] Optionally, according to the significance probability value of each initial independent variable under the current outlier threshold, the initial independent variables are screened from the initial feature set to obtain the feature set corresponding to the current outlier threshold. Specifically, in this embodiment, for each outlier threshold, linear regression processing is performed on each initial independent variable in the initial feature set to obtain the significance probability value of each initial independent variable under the current outlier threshold. Optionally, a threshold can be set to select the feature set corresponding to the current outlier threshold from the initial feature set.

[0043] Specifically, by adjusting the outlier threshold and repeating the feature screening operation at different thresholds, a series of feature sets can be formed, that is, multiple feature sets. Each feature set in the multiple feature sets corresponds to one of the multiple outlier thresholds respectively. For example, if there are 10 selected outlier thresholds, the number of corresponding multiple feature sets is also 10. Each feature set corresponds to a different outlier processing strategy. In this way, not only can it be understood which features are significant under different outlier tolerances, but also the impact of the outlier threshold on the final feature set and model performance can be evaluated.

[0044] After obtaining multiple feature sets, a target feature set can be selected from the multiple feature sets. Optionally, after obtaining multiple feature sets, according to the model performance metrics and business requirements, the most suitable feature set can be selected for the final model training. This process allows finding the best feature combination while considering outlier handling to build a reliable, robust, and efficient prediction model.

[0045] Through the embodiments provided in this application, according to the outlier threshold range and the threshold adjustment step size, multiple outlier thresholds are determined, and each outlier threshold in the multiple outlier thresholds is used as the current outlier threshold to perform the feature screening operation, obtaining multiple feature sets, where each feature set in the multiple feature sets corresponds to one outlier threshold in the multiple outlier thresholds respectively: According to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, each initial independent variable is sequentially subjected to linear regression processing to obtain the significance probability value of each initial independent variable under the current outlier threshold; According to the significance probability value of each initial independent variable under the current outlier threshold, initial independent variables are screened from the initial feature set to obtain the feature set corresponding to the current outlier threshold, so as to select the target feature set from the multiple feature sets by combining the correlation coefficient and the significance probability value, and train the prediction model, which can more accurately identify those independent variables that not only have a strong numerical correlation but also make a significant contribution to the prediction model in a statistical sense, so as to reduce the impact of outliers on the training of the prediction model, and solve the problem that the reliability of the prediction model obtained by training with these data is relatively low in the related art. Through the iterative screening process, the feature set that has a significant predictive effect on the dependent variable can be more accurately identified, and at the same time, non-significant features are excluded, thereby improving the interpretability and prediction efficiency of the model.

[0046] In an exemplary embodiment, the above method further includes: determining the average value of each initial independent variable in the initial feature set and the average value of the target dependent variable; according to the average value of each initial independent variable in the initial feature set and the average value of the target dependent variable, determining the covariance between each initial independent variable and the target dependent variable, the standard deviation of each initial independent variable, and the standard deviation of the target dependent variable; calculating the correlation coefficient between each initial independent variable and the target dependent variable according to the covariance between each initial independent variable and the target dependent variable, the standard deviation of each initial independent variable, and the standard deviation of the target dependent variable.

[0047] It should be noted that the correlation coefficient can be used to evaluate the strength and direction of the linear relationship between two variables. Specifically, it is used to evaluate the strength and direction of the linear correlation between the initial independent variable and the target dependent variable. Generally, the Pearson correlation coefficient can be used as the correlation coefficient. The average value of each initial independent variable can be used to represent the central tendency of the data of the same type included in each initial independent variable, so as to describe the central position of each initial independent variable. The average value of the target dependent variable can be used to represent the average value of the data of the same type included in the target dependent variable to describe the central position of the target dependent variable. Generally, the amount of data existing in the target dependent variable is greater than 1.

[0048] The covariance between each initial independent variable and the target dependent variable can be used to measure the degree of linear correlation between each initial independent variable and the target dependent variable. If the trends of the initial independent variable and the target dependent variable are the same, that is, they both increase or decrease simultaneously, the covariance is positive; if the trends of the initial independent variable and the target dependent variable are opposite, the covariance is negative. The magnitude and sign of the covariance value between each initial independent variable and the target dependent variable reflect the correlation and direction between each initial independent variable and the target dependent variable.

[0049] The standard deviation can be used to reflect the degree of dispersion of a set of values relative to their average value. The larger the standard deviation, the greater the volatility of the data. In this embodiment, the standard deviation of the initial independent variable is used to measure the degree of fluctuation of the data included in the initial independent variable; the standard deviation of the target dependent variable is used to measure the degree of fluctuation of the data included in the target dependent variable. The correlation coefficient can be a value between -1 and 1, which is used to describe the linear correlation between two variables. An absolute value close to 1 indicates a strong linear relationship between the two variables, and close to 0 indicates a weak linear relationship. A positive correlation means that the two variables change in the same direction, and a negative correlation means that one variable decreases when the other variable increases.

[0050] Through this embodiment, the correlation coefficients calculated between each initial independent variable and the target dependent variable can effectively reduce the interference of outliers on model training, improve the reliability of training data, and at the same time improve the interpretability and prediction accuracy of the model.

[0051] In an exemplary embodiment, according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, each initial independent variable is subjected to linear regression processing in turn to obtain the significance probability value of each initial independent variable under the current outlier threshold, including: according to the current outlier threshold, according to the arrangement order of the initial independent variables in the initial feature set, each initial independent variable is subjected to linear regression processing in turn to obtain the significance probability value of each initial independent variable under the current outlier threshold, wherein the arrangement order of the initial independent variables in the initial feature set is obtained by sorting the initial independent variables in the initial feature set according to the correlation coefficient between each initial independent variable and the target dependent variable.

[0052] It should be noted that the order of the initial independent variables in the initial feature set is determined according to the correlation coefficient between each initial independent variable and the dependent variable. The outlier threshold can be a parameter used to identify potential outliers. When the absolute value of the residual exceeds the set outlier threshold, the observation is considered an outlier. The setting of the threshold can adjust the sensitivity of the model to outliers. A too low threshold may eliminate too many normal data points, while a too high threshold may not effectively identify abnormal data.

[0053] Optionally, according to the current outlier threshold, in accordance with the arrangement order of the initial independent variables in the initial feature set, linear regression processing is performed on each initial independent variable in turn to obtain the significance probability value of each initial independent variable under the current outlier threshold. Specifically, for each initial independent variable subjected to linear regression processing, an adjustable loss function is determined by the current outlier threshold, and regression fitting is performed to obtain the regression coefficient and significance probability value of each initial independent variable under the current outlier threshold. Among them, the adjustable loss function gives a smaller weight to large residuals. In order to further perform feature screening, the loss value corresponding to the adjustable loss function under the current outlier threshold can be minimized, thereby obtaining the regression coefficient of each initial independent variable under the current outlier threshold and the significance probability value of each initial independent variable under the current outlier threshold.

[0054] Through this embodiment, linear regression processing is gradually performed on each initial independent variable in a manner of sorting by relevance, which eliminates the need for pre-identification and processing of outliers, simplifies the data preprocessing steps, reduces the complexity of early data cleaning, and is suitable for the feature screening process of massive independent variables.

[0055] In an exemplary embodiment, the method further includes: sorting the initial independent variables in the initial feature set in descending order of the absolute value of the correlation coefficient between each initial independent variable and the target dependent variable to obtain an arrangement order of the initial independent variables in the initial feature set.

[0056] It should be noted that the correlation coefficient can be a value between -1 and 1, which is used to describe the linear correlation between two variables. An absolute value close to 1 indicates a strong linear relationship between the two variables, and close to 0 indicates a weak linear relationship. A positive correlation means that the two variables change in the same direction, and a negative correlation means that one variable decreases when the other variable increases.

[0057] In the process of sorting the initial independent variables in the initial feature set, the absolute value of the correlation coefficient between each initial independent variable and the target dependent variable can be taken to obtain the absolute value of the correlation coefficient between each initial independent variable and the target dependent variable, and the initial independent variables in the initial feature set can be sorted in descending order according to the absolute value of the correlation coefficient between each initial independent variable and the target dependent variable, so as to perform linear regression processing in turn.

[0058] Through this embodiment, a high absolute value of the correlation coefficient represents a strong linear relationship between the independent variable and the target dependent variable. By first processing the initial independent variable corresponding to the high absolute value of the correlation coefficient, variables that have a significant impact on the prediction target can be preferentially identified and utilized, thereby improving the initial interpretability and prediction accuracy of the model.

[0059] In an exemplary embodiment, according to the current outlier threshold, in the order of arrangement of the initial independent variables in the initial feature set, linear regression processing is performed on each initial independent variable in turn to obtain the significance probability value of each initial independent variable under the current outlier threshold, including: constructing a linear regression model according to the current outlier threshold and a preset adjustable loss function; in the order of arrangement of the initial independent variables in the initial feature set, inputting each initial independent variable into the linear regression model in turn to obtain the significance probability value of each initial independent variable output by the linear regression model under the current outlier threshold; wherein, the significance probability value of each initial independent variable is determined based on the first independent variable corresponding to each initial independent variable and each initial independent variable, and the first independent variable corresponding to each initial independent variable is the initial independent variable whose significance probability value is less than the first significance threshold and is input into the linear regression model before each initial independent variable.

[0060] It should be noted that the preset adjustable loss function can be an index for measuring the prediction error of the model. In this embodiment, the loss function is designed to be adjustable. Especially when dealing with outliers, when the residual exceeds the outlier threshold, the loss function will automatically reduce the contribution of these outliers to the model fitting, so as to reduce the impact of outliers on the model parameter estimation. Specifically, the formula of the preset adjustable loss function can be:

[0061]

[0062] Among them, c is the outlier threshold, which is an adjustable constant. When |r| < c, the loss function is a smooth, non-linear function. As the residual r increases, the loss function gradually increases, but the increasing speed gradually slows down and finally reaches an upper limit. When |r| ≥ c, the value of the loss function is fixed.

[0063] Optionally, according to the current outlier threshold and the preset adjustable loss function, a linear regression model can be constructed, so that the constructed linear regression model will automatically adapt to the existence of outliers during the fitting stage and reduce the impact of outliers on model fitting. For different outlier thresholds, a linear regression model corresponding to the outlier threshold can be constructed based on the outlier threshold and the preset adjustable loss function.

[0064] Specifically, the form of the linear regression model can be expressed by formula (2):

[0065] y = βX + ε; (2)

[0066] Among them, X is the independent variable, y is the dependent variable, β is the influence coefficient, also known as the regression coefficient, and ε is the preset acceptable residual.

[0067] In specific practice, the residual can be expressed by formula (3):

[0068]

[0069] Among them, r i represents the residual of the i-th observation value; y i represents the actual value of the target dependent variable of the i-th observation value, which is the value that is expected to be predicted by the model, represents the estimated value of the regression coefficient of the initial independent variable, which determines the weight of the initial independent variable X i when predicting the target dependent variable; X i represents the values of all independent variables of the i-th observation value, usually represented in vector form. Generally, X i can include one or more data, specifically depending on the number of features included in the model.

[0070] After constructing the linear regression model, linear regression processing can be performed on each initial independent variable in turn according to the arrangement order of the initial independent variables in the sorted initial feature set. Specifically, according to the arrangement order of the initial independent variables in the sorted initial feature set, each initial independent variable is input into the linear regression model corresponding to the current outlier threshold in turn for fitting processing, and the significance probability value of each initial independent variable under the current outlier threshold is calculated. The significance probability value of each initial independent variable under the current outlier threshold can be the significance probability value of each initial independent variable newly input into the linear regression model under the current outlier threshold. The significance probability value of each initial independent variable is determined based on the first independent variable corresponding to each initial independent variable and each initial independent variable. The first independent variable corresponding to each initial independent variable is the initial independent variable whose significance probability value is less than the first significance threshold and is input into the linear regression model before each initial independent variable.

[0071] The significance probability value can be used to quantify the degree of inconsistency between the observed data and the hypothesized model to test whether an independent variable significantly affects the target variable. In this embodiment, the significance probability value can be calculated by the linear regression model, and the significance probability value can be used to determine whether the initial independent variable makes a significant contribution to the target dependent variable under the current outlier threshold setting. The first significance threshold can be a preset threshold based on experience and is used to judge the significance of the initial independent variable. If the significance probability value (P-value) of the initial independent variable is less than the first significance threshold, it is considered that the initial independent variable with a significance probability value less than the first significance threshold has a significant impact on the target dependent variable and can be retained for subsequent model construction. If the significance probability value of the initial independent variable is greater than or equal to the first significance threshold, it can be determined that adding this initial independent variable does not significantly improve the predictive ability of the model, so it is not retained, and the next initial independent variable is continued to be processed until all initial independent variables have been evaluated once, or until the significance test no longer identifies new significant features. Finally, a linear regression model containing all the initial independent variables that pass the significance test is obtained.

[0072] Through this embodiment, using the preset adjustable loss function can reduce the impact of outliers on the linear regression model, making the linear regression model more stable and reliable when dealing with imperfect data. By gradually adding the initial independent variables in the order of relevance, it is possible to give priority to those initial independent variables that are most closely related to the target dependent variable.

[0073] In an exemplary embodiment, according to the significance probability value of each initial independent variable under the current outlier threshold, initial independent variables are screened from the initial feature set to obtain a feature set corresponding to the current outlier threshold, including: in the case where there are initial independent variables with significance probability values less than the second significance threshold in the initial feature set, based on the initial independent variables with significance probability values less than the second significance threshold in the initial feature set, a feature set corresponding to the current outlier threshold is constituted.

[0074] It should be noted that the threshold is a key parameter of the loss function, which affects the sensitivity to outliers when fitting a linear regression model. The significance probability value can be an index used to judge whether the influence of the initial independent variable on the target dependent variable is significant. Among them, the smaller the significance probability value, the less likely it is that the relationship between the initial independent variable and the target dependent variable is caused by random variation. Therefore, the contribution of the initial independent variable to the linear regression model is more significant. The second significance threshold can be used to judge whether the significance probability value of the independent variable is sufficient to regard it as a significant feature. Generally, initial independent variables with significance probability values less than the second significance threshold are selected to construct the feature set.

[0075] Among them, the second significance threshold can be set based on actual needs. The magnitudes of the first significance threshold and the second significance threshold can be the same or different, and this application does not make a limitation here.

[0076] In specific practice, after determining the feature set corresponding to the current outlier threshold, a preset algorithm can be used to minimize the loss value of the feature set corresponding to the current outlier threshold when constructing a linear regression model based on the current outlier threshold, and according to the loss value, the fitting value of the linear regression model constructed based on the current outlier threshold is determined. Among them, the preset algorithm can be the gradient descent method. Specifically: by using formula (4) to minimize the following loss value L, the fitting value of the regression model, that is, the regression coefficient, can be obtained:

[0077]

[0078] Among them, L is the loss value, r i represents the residual of the i-th observation value; y i represents the actual value of the target dependent variable of the i-th observation value, which is the value that is hoped to be predicted by the model. represents the estimated value of the regression coefficient of the initial independent variable, which determines the weight of the initial independent variable X i when predicting the target dependent variable; X i represents the values of all independent variables of the i-th observation value, usually represented in vector form. Generally, X i can include one or more data, specifically depending on the number of features included in the model, and N represents the total number of observation values in the dataset, that is, the sample size.

[0079] In this embodiment, by setting the outlier threshold, it is less sensitive to outliers in the data, so that noise data can be better processed, and the robustness of the prediction model is improved. Using the second significance threshold for feature screening ensures that the linear regression model only includes independent variables that make significant contributions to the prediction, avoiding overfitting and model redundancy, thereby improving the prediction accuracy. Moreover, by only including independent variables that make significant contributions to the prediction, the selected feature set can more accurately reflect the relationship between the independent variables and the target variable, thus constructing a more accurate and stable prediction model.

[0080] In an exemplary embodiment, step S206 includes: obtaining the loss value corresponding to each outlier threshold; determining the feature set corresponding to the outlier threshold with the smallest corresponding loss value among multiple feature sets as the target feature set.

[0081] Specifically, the loss value refers to the gap between the model prediction result and the actual result calculated by the loss function during the model training process. The design goal of the loss function is to make the model prediction as close to the true value as possible. The smaller the loss value, the smaller the prediction error of the model and the higher the fitting degree. After determining multiple feature sets (the feature set corresponding to each outlier threshold), the loss value corresponding to each outlier threshold can be calculated according to the feature set corresponding to each outlier threshold. By comparing the loss values of the linear regression model under different outlier thresholds, select the feature set corresponding to the outlier threshold with the smallest loss value as the target feature set.

[0082] In this embodiment, by comparing the loss values under different outlier thresholds, a threshold that minimizes the model prediction error can be found, thereby constructing an optimal prediction model. Automatically adjusting the sensitivity to outliers during the model construction process avoids the negative impact of outliers on the model performance and improves the robustness of the model. Under the influence of the outlier threshold, features that make significant contributions to the target variable are screened out, which helps to construct a model that is both accurate and concise, avoiding feature redundancy and overfitting.

[0083] The following explains the feature screening method in the embodiments of the present application with reference to optional examples. Figure 3 is a schematic flowchart of the feature screening method in this optional example, as Figure 3 shown, the process of this feature screening method may include the following steps:

[0084] Step S302, obtain the current outlier threshold;

[0085] Specifically, obtain the initial feature set and the target dependent variable. Determine multiple outlier thresholds according to the outlier threshold range and the threshold adjustment step size, and use each outlier threshold among the multiple outlier thresholds as the current outlier threshold to perform feature screening operations in sequence.

[0086] Step S304, obtain the arrangement order of the initial independent variables in the initial feature set;

[0087] Specifically, sort the initial independent variables in the initial feature set from largest to smallest according to the absolute value of the correlation coefficient between the target dependent variable and the initial independent variables in the initial feature set, and obtain the arrangement order of the initial independent variables in the initial feature set.

[0088] Step S306, determine whether the number of remaining initial independent variables is greater than 0;

[0089] Specifically, under the current outlier threshold, the number of initial independent variables in the initial feature set that have not been fitted, that is, the number of remaining initial independent variables.

[0090] If the number of remaining initial independent variables is greater than 0, jump to step S308; if the number of remaining initial independent variables is greater than 0, jump to step S310.

[0091] Step S308, obtain the current initial independent variable and fit the current initial independent variable to obtain the significance probability value of the current initial independent variable;

[0092] Step S310, traversal completed;

[0093] Step S312, determine whether the significance probability value meets the threshold;

[0094] Specifically, if the significance probability value meets the threshold, that is, when the significance probability value is less than the significance threshold, then jump to step S314; when the significance probability value is greater than or equal to the significance threshold, then jump to step S316.

[0095] Step S314, add the significant feature set corresponding to the current outlier threshold, and perform regression fitting on the significant feature set to obtain the current loss value and the fitted value;

[0096] Specifically, after performing regression fitting on the significant feature set, jump to step S306.

[0097] Step S316, add the non-significant feature set corresponding to the current outlier threshold.

[0098] Through this optional example, by setting different outlier thresholds, the sensitivity to outliers can be automatically adjusted, avoiding the negative impact of outliers on the model performance and improving the robustness of the model. Based on the correlation coefficient sorting and significance test, this embodiment can screen out the features that contribute significantly to the target variable from the initial feature set, avoiding the influence of noise features on the model and improving the prediction accuracy of the model. By adopting the step-by-step feature screening method, compared with considering all possible feature combinations at once, the computational complexity is greatly reduced and the efficiency of model construction is improved.

[0099] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0100] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM (Read-Only Memory), RAM (Random Access Memory), magnetic disk, optical disk), and includes several instructions to enable a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of this application.

[0101] According to another aspect of the embodiments of this application, a feature screening device is further provided. This feature screening device can be used to implement the feature screening method provided in the above embodiments, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0102] Figure 4 is a structural block diagram of an optional feature screening device according to the embodiments of this application. As Figure 4 shown in, this feature screening device includes:

[0103] An acquisition unit 402, configured to acquire an initial feature set and a target dependent variable, wherein each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable.

[0104] A screening unit 404, configured to determine a plurality of outlier thresholds according to an outlier threshold range and a threshold adjustment step size, and use each outlier threshold in the plurality of outlier thresholds as a current outlier threshold to perform the following feature screening operations to obtain a plurality of feature sets, wherein each feature set in the plurality of feature sets corresponds to one outlier threshold in the plurality of outlier thresholds respectively: perform linear regression processing on each initial independent variable in turn according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold to obtain a significance probability value of each initial independent variable under the current outlier threshold; screen the initial independent variables from the initial feature set according to the significance probability value of each initial independent variable under the current outlier threshold to obtain a feature set corresponding to the current outlier threshold;

[0105] A selection unit 406, configured to select a target feature set from the plurality of feature sets, wherein the target feature set is used to train a prediction model corresponding to the target dependent variable.

[0106] It should be noted that the acquisition unit 402 in this embodiment can be used to execute the above step S202, the screening unit 404 in this embodiment can be used to execute the above step S204, and the selection unit 406 in this embodiment can be used to execute the above step S206.

[0107] Through the embodiments provided in this application, according to the outlier threshold range and the threshold adjustment step size, multiple outlier thresholds are determined, and each outlier threshold among the multiple outlier thresholds is used as the current outlier threshold to perform a feature screening operation, obtaining multiple feature sets, where each feature set among the multiple feature sets corresponds to one outlier threshold among the multiple outlier thresholds: According to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold, each initial independent variable is sequentially subjected to linear regression processing to obtain the significance probability value of each initial independent variable under the current outlier threshold; According to the significance probability value of each initial independent variable under the current outlier threshold, initial independent variables are screened from the initial feature set to obtain the feature set corresponding to the current outlier threshold. By combining the use of the correlation coefficient and the significance probability value, a target feature set is selected from the multiple feature sets to train the prediction model, which can more accurately identify those independent variables that not only have a strong numerical correlation but also make a significant contribution to the prediction model in terms of statistical significance, so as to reduce the impact of outliers on the training of the prediction model, solve the problem that the reliability of the prediction model obtained by training using these data is relatively low in the related art. Through the iterative screening process, the feature set that has a significant predictive effect on the dependent variable can be more accurately identified, and at the same time, non-significant features are excluded, thereby improving the interpretability and prediction efficiency of the model.

[0108] In an exemplary embodiment, the above device further includes: a coefficient calculation unit, configured to determine the average value of each initial independent variable in the initial feature set and the average value of the target dependent variable; According to the average value of each initial independent variable in the initial feature set and the average value of the target dependent variable, determine the covariance between each initial independent variable and the target dependent variable, the standard deviation of each initial independent variable, and the standard deviation of the target dependent variable; According to the covariance between each initial independent variable and the target dependent variable, the standard deviation of each initial independent variable, and the standard deviation of the target dependent variable, calculate the correlation coefficient between each initial independent variable and the target dependent variable.

[0109] In an exemplary embodiment, the screening unit 404 is further configured to, according to the current outlier threshold, sequentially perform linear regression processing on each initial independent variable in the order of arrangement of the initial independent variables in the initial feature set, to obtain the significance probability value of each initial independent variable under the current outlier threshold, where the order of arrangement of the initial independent variables in the initial feature set is obtained by sorting the initial independent variables in the initial feature set according to the correlation coefficient between each initial independent variable and the target dependent variable.

[0110] In an exemplary embodiment, the above-mentioned device further includes: a sorting unit, configured to sort the initial independent variables in the initial feature set in descending order of the absolute value of the correlation coefficient between each initial independent variable and the target dependent variable, so as to obtain the arrangement order of the initial independent variables in the initial feature set.

[0111] In an exemplary embodiment, the screening unit 404 is further configured to: construct a linear regression model according to the current outlier threshold and a preset adjustable loss function; sequentially input each initial independent variable in the initial feature set into the linear regression model according to the arrangement order of the initial independent variables in the initial feature set, so as to obtain the significance probability value of each initial independent variable output by the linear regression model under the current outlier threshold; wherein, the significance probability value of each initial independent variable is determined based on the first independent variable corresponding to each initial independent variable and each initial independent variable, and the first independent variable corresponding to each initial independent variable is the initial independent variable whose significance probability value is less than the first significance threshold and is input into the linear regression model before each initial independent variable.

[0112] In an exemplary embodiment, the screening unit 404 is further configured to: in the case that there is an initial independent variable in the initial feature set whose significance probability value is less than the second significance threshold, form a feature set corresponding to the current outlier threshold based on the initial independent variables in the initial feature set whose significance probability values are less than the second significance threshold.

[0113] In an exemplary embodiment, the selection unit 406 is further configured to: obtain the loss value corresponding to each outlier threshold; determine the feature set corresponding to the outlier threshold with the smallest corresponding loss value among the multiple feature sets as the target feature set.

[0114] It should be noted that the above-mentioned various modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to this: all the above-mentioned modules are located in the same processor; or, the above-mentioned various modules are respectively located in different processors in any combination form.

[0115] According to another aspect of the embodiments of the present application, there is provided a computer-readable storage medium. The computer-readable storage medium includes a stored program, wherein the program executes the steps in any one of the above method embodiments when running.

[0116] In an exemplary embodiment, the above-mentioned computer-readable storage medium may include, but is not limited to: various media such as USB flash drives, ROMs, RAMs, mobile hard disks, magnetic disks or optical discs that can store computer programs.

[0117] According to another aspect of the embodiments of the present application, an electronic device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor is configured to execute the steps in any of the above method embodiments through the computer program. In an exemplary embodiment, the above electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0118] Specific examples in this embodiment may refer to the examples described in the above embodiments and exemplary embodiments, and will not be elaborated herein.

[0119] According to another aspect of the embodiments of the present application, a computer program product is further provided. The computer program product includes computer programs / instructions, and the computer programs / instructions contain program codes for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 509 and / or installed from the removable medium 511. When the computer program is executed by the central processing unit 501, various functions provided by the embodiments of the present application are executed. The above serial numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0120] Figure 5 Schematically shows a block diagram of a computer system of an electronic device for implementing the embodiments of the present application. As Figure 5 shown, the computer system 500 includes a CPU (Central Processing Unit) 501, which can execute various appropriate actions and processes according to the program stored in the ROM 502 or the program loaded from the storage part 508 into the RAM 503. In the random access memory 503, various programs and data required for system operation are also stored. The central processing unit 501, the read-only memory 502, and the random access memory 503 are connected to each other through a bus 504. The I / O (Input / Output) interface 505 is also connected to the bus 504.

[0121] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including such as a CRT (Cathode Ray Tube), an LCD (Liquid Crystal Display), etc. and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a local area network card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the input / output interface 505 as required. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as required so that a computer program read from it can be installed into the storage section 508 as required.

[0122] Specifically, according to an embodiment of the present application, the processes described in each method flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through the communication section 509, and / or installed from the removable medium 511. When the computer program is executed by the central processing unit 501, various functions defined in the system of the present application are executed.

[0123] It should be noted that Figure 5 The computer system 500 of the electronic device shown is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0124] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present application can be implemented by a general computing device. They can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order from here, or they can be separately made into individual integrated circuit modules, or multiple modules or steps among them can be made into a single integrated circuit module to implement. In this way, the present application is not limited to any specific combination of hardware and software.

[0125] The above is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the principle of the present application shall be included in the protection scope of the present application.

Claims

1. A feature screening method, characterized in that, Including: Obtain an initial feature set and a target dependent variable, where each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable; According to an outlier threshold range and a threshold adjustment step size, determine a plurality of outlier thresholds, and use each outlier threshold in the plurality of outlier thresholds as a current outlier threshold to perform the following feature screening operations to obtain a plurality of feature sets, where each feature set in the plurality of feature sets corresponds to one outlier threshold in the plurality of outlier thresholds respectively: perform linear regression processing on each initial independent variable in turn according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold to obtain the significance probability value of each initial independent variable under the current outlier threshold; screen the initial independent variables from the initial feature set according to the significance probability value of each initial independent variable under the current outlier threshold to obtain the feature set corresponding to the current outlier threshold; Select a target feature set from the plurality of feature sets, where the target feature set is used to train a prediction model corresponding to the target dependent variable.

2. The method according to claim 1, wherein The method further includes: Determine the average value of each initial independent variable in the initial feature set and the average value of the target dependent variable; According to the average value of each initial independent variable in the initial feature set and the average value of the target dependent variable, determine the covariance between each initial independent variable and the target dependent variable, the standard deviation of each initial independent variable, and the standard deviation of the target dependent variable; Calculate the correlation coefficient between each initial independent variable and the target dependent variable according to the covariance between each initial independent variable and the target dependent variable, the standard deviation of each initial independent variable, and the standard deviation of the target dependent variable.

3. The method according to claim 1, wherein The performing linear regression processing on each initial independent variable in turn according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold to obtain the significance probability value of each initial independent variable under the current outlier threshold includes: According to the current outlier threshold, perform linear regression processing on each initial independent variable in turn according to the arrangement order of the initial independent variables in the initial feature set to obtain the significance probability value of each initial independent variable under the current outlier threshold, where the arrangement order of the initial independent variables in the initial feature set is obtained by sorting the initial independent variables in the initial feature set according to the correlation coefficient between each initial independent variable and the target dependent variable.

4. The method according to claim 3, characterized in that, The method further includes: Sort the initial independent variables in the initial feature set in descending order according to the absolute value of the correlation coefficient between each initial independent variable and the target dependent variable to obtain the arrangement order of the initial independent variables in the initial feature set.

5. The method according to claim 3, characterized in that, Performing linear regression processing on each of the initial independent variables in the order of arrangement of the initial independent variables in the initial feature set according to the current outlier threshold to obtain the significance probability value of each initial independent variable under the current outlier threshold, including: Constructing a linear regression model according to the current outlier threshold and a preset adjustable loss function; Sequentially inputting each of the initial independent variables into the linear regression model in the order of arrangement of the initial independent variables in the initial feature set to obtain the significance probability value of each initial independent variable output by the linear regression model under the current outlier threshold; Wherein, the significance probability value of each initial independent variable is determined based on the first independent variable corresponding to each initial independent variable and each initial independent variable, and the first independent variable corresponding to each initial independent variable is the initial independent variable whose significance probability value is less than the first significance threshold and is input into the linear regression model before each initial independent variable.

6. The method according to claim 5, wherein Screening initial independent variables from the initial feature set according to the significance probability value of each initial independent variable under the current outlier threshold to obtain the feature set corresponding to the current outlier threshold, including: In the case that there are initial independent variables in the initial feature set whose significance probability values are less than the second significance threshold, constructing a feature set corresponding to the current outlier threshold based on the initial independent variables in the initial feature set whose significance probability values are less than the second significance threshold.

7. The method according to claim 6, characterized in that, Selecting a target feature set from the multiple feature sets, including: Obtaining the loss value corresponding to each outlier threshold; Determining the feature set corresponding to the outlier threshold with the smallest corresponding loss value among the multiple feature sets as the target feature set.

8. A feature screening device, characterized in that, Including: An acquisition unit for acquiring an initial feature set and a target dependent variable, wherein each initial independent variable in the initial feature set is a candidate variable for predicting the target dependent variable; A screening unit for determining a plurality of outlier thresholds according to an outlier threshold range and a threshold adjustment step size, and taking each outlier threshold in the plurality of outlier thresholds as the current outlier threshold, and performing the following feature screening operations to obtain a plurality of feature sets, wherein each feature set in the plurality of feature sets respectively corresponds to one outlier threshold in the plurality of outlier thresholds: Performing linear regression processing on each of the initial independent variables in sequence according to the correlation coefficient between each initial independent variable and the target dependent variable and the current outlier threshold to obtain the significance probability value of each initial independent variable under the current outlier threshold; Screening initial independent variables from the initial feature set according to the significance probability value of each initial independent variable under the current outlier threshold to obtain the feature set corresponding to the current outlier threshold; A selection unit for selecting a target feature set from the multiple feature sets, wherein the target feature set is used to train a prediction model corresponding to the target dependent variable.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, wherein when the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 7.

Images