Feature screening methods, apparatuses, devices, and media

Abstract

The application provides a feature screening method, device, equipment and medium, relates to big data analysis, artificial intelligence and feature engineering technology. The method comprises the following steps: calculating the correlation degree index of the feature data sequence of each feature and the feature data sequence of other features to obtain a correlation degree index table; selecting the features i and j corresponding to the correlation degree index with the highest absolute value from the correlation degree index table; if the correlation degree index with the highest absolute value exceeds a preset threshold, determining the feature to be deleted from the correlation degree index according to the correlation of the features i and j with other features; deleting the feature to be deleted to obtain a new feature data table; returning to execute the step of calculating the correlation degree index of the feature data sequence of each feature and the feature data sequence of other features until the correlation degree index with the highest absolute value in the correlation degree index table of the new feature data table does not exceed the preset threshold, and taking the remaining features at this time as target features obtained through screening.

Description

Technical Field

[0001] This application relates to the fields of big data analysis, artificial intelligence and feature engineering technology, and in particular to a feature screening method, apparatus, device and medium. Background Technology

[0002] Feature engineering is the foundation of modern artificial intelligence models, especially for regression and classification problems. The quality of the feature variable data used by the model directly determines the model's predictive performance.

[0003] Traditional feature engineering techniques rely heavily on the modeler's own modeling experience and lack the technology to intelligently filter features, thus affecting modeling efficiency. Summary of the Invention

[0004] This application provides a feature screening method, apparatus, device, and medium to solve the problem that intelligent feature screening cannot be achieved in the prior art.

[0005] Firstly, this application provides a feature selection method, including:

[0006] Obtain a feature data table of n samples under m dimensions, and take the n feature data under any dimension as a feature data sequence;

[0007] Calculate the correlation index between the feature data sequence of each feature and the feature data sequences of other features to obtain the correlation index table;

[0008] Select feature i and feature j corresponding to the correlation index with the highest absolute value from the correlation index table;

[0009] If the correlation index with the highest absolute value exceeds the preset threshold, then the feature to be deleted is determined from feature i and feature j based on the correlation between feature i and feature j and other features.

[0010] The feature to be deleted and its feature data sequence are deleted to obtain a new feature data table;

[0011] For the new feature data table, return to the step of calculating the correlation index between the feature data sequence of each feature and the feature data sequence of other features, until the correlation index with the highest absolute value in the correlation index table of the new feature data table does not exceed the preset threshold, and use the remaining features at this time as the target features obtained by filtering.

[0012] Secondly, this application provides a feature screening device, comprising:

[0013] The feature data acquisition module is used to acquire feature data tables of n samples under m-dimensional features, and to take the n feature data under any dimension as a feature data sequence.

[0014] The correlation index calculation module is used to calculate the correlation index between the feature data sequence of each feature and the feature data sequence of other features, and obtain the correlation index table.

[0015] The selection module is used to select feature i and feature j corresponding to the correlation index with the highest absolute value from the correlation index table;

[0016] The feature to be deleted module is used to determine the feature to be deleted from feature i and feature j based on the correlation between feature i and feature j and other features if the correlation index with the highest absolute value exceeds a preset threshold.

[0017] The feature deletion module is used to delete the feature to be deleted and its feature data sequence to obtain a new feature data table;

[0018] The loop processing module is used to return to the step of calculating the correlation index between the feature data sequence of each feature and the feature data sequence of other features for the new feature data table, until the correlation index with the highest absolute value in the correlation index table of the new feature data table does not exceed the preset threshold, and the remaining features at this time are used as the target features obtained by filtering.

[0019] Thirdly, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the feature screening method as described in any of the embodiments of this application.

[0020] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the feature selection method as described in any of the embodiments of this application.

[0021] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the feature selection method as described in any of the embodiments of this application.

[0022] The feature filtering method, apparatus, device, and medium provided in this application measure the correlation between different features using a correlation index. First, the correlation index between any two features is calculated. Then, the correlation index with the highest absolute value is selected, thus determining the corresponding two features, feature i and feature j. Next, the overall correlation between feature i and feature j and other features is examined, and features to be deleted are determined from feature i and feature j based on the overall correlation. This process is repeated iteratively based on the deleted features until no two features have an absolute correlation index exceeding a preset threshold, thus obtaining the final feature set. This method intelligently filters features, resulting in a more concise feature set that contains more information, greatly improving the effectiveness of feature filtering. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a feature filtering method provided in an embodiment of this application;

[0025] Figure 2 A flowchart illustrating the process of determining features to be deleted in the feature filtering method provided in this application embodiment;

[0026] Figure 3 A flowchart illustrating the method for modifying features in the feature filtering method provided in this application embodiment;

[0027] Figure 4 This is a schematic diagram of the structure of a feature screening device provided in an embodiment of this application;

[0028] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0029] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0030] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0031] It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first" and "second" are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. The acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant national laws and regulations. It should be noted that in the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used the relevant content of that solution.

[0032] Figure 1 This is a flowchart illustrating a feature filtering method provided in an embodiment of this application. This embodiment is applicable to filtering feature data used for training a model. The method can be executed by a feature filtering device, which can be implemented in software and / or hardware, preferably configured in an electronic device, such as a computer or server. Figure 1 As shown, the method specifically includes:

[0033] S101. Obtain the feature data table of n samples under m dimensions, and take the n feature data under any dimension as the feature data sequence.

[0034] S102. Calculate the correlation index between the feature data sequence of each feature and the feature data sequence of other features to obtain the correlation index table.

[0035] S103. Select feature i and feature j corresponding to the correlation index with the highest absolute value from the correlation index table.

[0036] S104. Determine whether the correlation index with the highest absolute value exceeds the preset threshold. If yes, proceed to S105; otherwise, proceed to S107.

[0037] S105. Based on the correlation between feature i and feature j and other features, determine the features to be deleted from feature i and feature j.

[0038] S106. Delete the feature to be deleted and its feature data sequence to obtain a new feature data table, and return to execute S102 for the new feature data table.

[0039] S107. Use the remaining features at this point as the target features obtained through filtering.

[0040] This application embodiment filters features based on the correlation between features. Specifically, it first obtains a feature data table of n samples under m dimensions of features, and then uses the n feature data under any dimension of features as a feature data sequence. For example, the feature data table is represented as follows:

[0041] serial number Feature 1 Feature 2 Feature 3 Feature 4 Feature 5 …… Feature m-1 Feature m 1 x 11 ]]> x 12 ]]> x 13 ]]> x 14 ]]> x 15 ]]> …… x 1(m-1) ]]> x 1m ]]> 2 x 21 ]]> x 22 ]]> x 23 ]]> x 24 ]]> x 25 ]]> …… <![CDATA[x 2(m-1) ]]> <![CDATA[x 2m ]]> 3 <![CDATA[x 31 ]]> <![CDATA[x 32 ]]> <![CDATA[x 33 ]]> <![CDATA[x 34 ]]> <![CDATA[x 35 ]]> …… x 3(m-1) ]]> <![CDATA[x 3m ]]> 4 <![CDATA[x 41 ]]> <![CDATA[x 42 ]]> <![CDATA[x 43 ]]> <![CDATA[x 44 ]]> <![CDATA[x 45 ]]> …… <![CDATA[x 4(m-1) ]]> <![CDATA[x 4m ]]> …… …… …… …… …… …… …… …… …… n-1 <![CDATA[x (n-1)1 ]]> <![CDATA[x (n-1)2 ]]> <![CDATA[x (n-1)3 ]]> <![CDATA[x (n-1)4 ]]> <![CDATA[x (n-1)5 ]]> …… <![CDATA[x (n-1)(m-1) ]]> <![CDATA[x (n-1)m ]]> n <![CDATA[x n1 ]]> <![CDATA[x n2 ]]> <![CDATA[x n3 ]]> <![CDATA[x n4 ]]> <![CDATA[x n5 ]]> …… <![CDATA[x n(m-1) ]]> <![CDATA[x nm ]]>

[0042] Where n represents the number of observed samples, and m represents the number of feature dimensions. Each row in the table represents the feature data of an observed sample under each feature dimension, and each column represents the feature data sequence of a feature variable. For example, suppose a financial institution wants to predict customers' willingness to purchase new financial products and needs to perform machine learning modeling analysis. Currently, there are 10,000 observed customer samples, corresponding to n = 10,000 in the table above, and each customer's data contains 10 features, corresponding to m = 10 in the table above.

[0043] Different features refer to describing observed samples from different dimensions or perspectives. However, feature variables of different dimensions may produce the same changes. The more such feature variables there are, the more it will increase the complexity of the model, rather than improve the amount of data information the model can acquire. Therefore, it is necessary to intelligently filter the features of m dimensions. The target features obtained after filtering can then be used to train the current target model, thereby improving the efficiency of model training.

[0044] First, calculate the correlation index between the feature data sequence of each feature and the feature data sequences of other features to obtain a correlation index table. For example, correlation coefficients such as Pearson correlation coefficient, Spearman correlation coefficient, or Kendall rank correlation coefficient can all be used as correlation indexes to measure the degree of correlation between features. An exemplary correlation index table that can represent the pairwise correlation between features can be constructed as follows:

[0045] 1 2 3 …… D 1 <![CDATA[ρ 11 ]]> <![CDATA[ρ 12 ]]> <![CDATA[ρ 13 ]]> …… <![CDATA[ρ 1D ]]> 2 <![CDATA[ρ 21 ]]> <![CDATA[ρ 22 ]]> <![CDATA[ρ 23 ]]> …… <![CDATA[ρ 2D ]]> 3 <![CDATA[ρ 31 ]]> <![CDATA[ρ 32 ]]> <![CDATA[ρ 33 ]]> …… <![CDATA[ρ 3D ]]> …… …… …… …… …… …… D <![CDATA[ρ D1 ]]> <![CDATA[ρ D2 ]]> <![CDATA[ρ D3 ]]> …… <![CDATA[ρ DD ]]>

[0046] Where, ρ ij Let ρ represent the correlation coefficient between the feature data sequences of the i-th feature and the j-th feature, and let ρ be the correlation coefficient between them. ij =ρ ji For ρ ii ,i=1,…,D (D represents the number of features), which represents the correlation coefficient between feature i and itself, and is fixed at 1.

[0047] Next, select features i and j from the correlation index table that correspond to the highest absolute value of the correlation index. In the correlation index table, if the absolute value of the correlation index of the feature data sequences of features i and j is the highest, it means that the correlation between features i and j is the highest compared to any other pairwise features. Therefore, if the correlation index of features i and j does not exceed the preset threshold, then the correlation index of any other pairwise features must also not exceed the threshold, and therefore, no features need to be deleted. Conversely, if the correlation index of features i and j exceeds the preset threshold, then one of them needs to be deleted to achieve the purpose of simplifying features.

[0048] In this embodiment, the feature to be deleted is determined from features i and j based on their respective correlations with other features. That is, if one of features i and j has a higher overall correlation with other features, it indicates that the information of that feature is more easily expressed by other features, and therefore that feature can be deleted. The overall correlation between features i and j and other features can also be measured using correlation coefficients, and this application does not impose any limitations on this method.

[0049] After deleting the feature to be deleted and its feature data sequence, a new feature data table is obtained. Then, for the new feature data table, the process returns to S102. That is, if there are two features in the new feature data table with the highest absolute value of their correlation index that exceeds the preset threshold, then one of them needs to be deleted. This process is repeated iteratively until the correlation index with the highest absolute value in a new feature data table does not exceed the preset threshold. The remaining features at this point are then used as the target features for the selection.

[0050] The feature filtering method, apparatus, device, and medium provided in this application measure the correlation between different features using a correlation index. First, the correlation index between any two features is calculated. Then, the correlation index with the highest absolute value is selected, thus determining the corresponding two features, feature i and feature j. Next, the overall correlation between feature i and feature j and other features is examined, and features to be deleted are determined from feature i and feature j based on the overall correlation. This process is repeated based on the deleted features until no two features have an absolute correlation index exceeding a preset threshold, resulting in the final feature set. This method intelligently filters features, resulting in a more concise feature set containing more information, greatly improving the effectiveness of feature filtering.

[0051] In one implementation, the correlation index includes at least one of the following: Pearson correlation coefficient, Spearman correlation coefficient, Kendall rank correlation coefficient, and synchronous rate of change, wherein the synchronous rate of change is used to observe the phenomenon of synchronous changes in data over a certain time span.

[0052] Specifically, different correlation coefficients reflect different aspects of correlation. For example, the Pearson correlation coefficient is a linear correlation coefficient, reflecting linear correlation, while the Spearman correlation coefficient and Kendall rank correlation coefficient are both rank correlation coefficients, reflecting whether the order of the two features is consistent, and thus determining whether the two features grow or decline synchronously.

[0053] In this embodiment, the three correlation coefficients and synchronous change rate mentioned above can be considered simultaneously. This allows for the measurement of the correlation between different features from multiple angles and aspects, making the correlation analysis results more accurate and enabling the discovery of highly correlated features that are not easily detected, thus facilitating better feature selection. Specifically, one of the three correlation coefficients and synchronous change rate can be selected as the correlation index for feature calculation, and a first round of selection is performed based on the correlation index table according to the feature selection method described in the above embodiment. Then, for the feature set obtained in the first round of selection, another correlation index is selected for calculation, and a second round of selection is performed using the same selection method. This process is repeated for a total of four rounds of selection to obtain the final target feature set. It should be noted that this embodiment does not impose any limitation on the order of the correlation indices selected in each round of selection.

[0054] Furthermore, the synchronous change rate in this embodiment is a different evaluation metric system from the correlation coefficient, based on the consistency of synchronous changes of different features. Since the correlation coefficient between features only compares data from the same period, but the correlation between many data points often has a lag, the correlation coefficient cannot detect this. For example, data from two sets of features X and Y collected over the past n days (x1, y1), (x2, y2), ..., (x... n ,y n In this case, the subscripts of x and y represent the dates. If a correlation coefficient is used, the analysis process involves combining the feature X data x1 and feature Y data y1 from the first day into a data pair (x1, y1), and then incorporating them into the correlation analysis. However, in reality, the data of feature X and feature Y on the same day may not be correlated, but the data from a day apart may be highly correlated. That is, x1 and y1 are unrelated, but x1 and y2 are highly correlated. In this case, either correlation coefficient will incorrectly determine that feature X and feature Y are uncorrelated.

[0055] The synchronous change rate in this embodiment is not analyzed based on the correlation between changes in data at a fixed single point in time. Instead, it is determined by observing whether there is a high degree of synchronous change phenomenon in the data over a certain time span, as detailed below:

[0056] Suppose any two sets of data for features X and Y are (x1, y1), (x2, y2), ..., (x... n ,y n ), can be used to construct synchronous rates of change:

[0057]

[0058] dx i =(x i -x i-1 ) / x i-1 i = 2, ..., n

[0059] dy i =(y i -y i-1 ) / y i-1 i = 2, ..., n

[0060] Among them, tc XY dx represents the synchronous rate of change of feature X and feature Y. i and dy i Let x represent the data in the feature data sequence of feature X and feature Y respectively. i Data y i The percentage increase;

[0061] For synchronous characteristic functions, when dx i with dy i dy i-1 dy i+1 If any two values ​​of the same sign are compared, and their growth rates differ by no more than a target multiple (e.g., 1.5 times), then I is 1; otherwise, I is 0. For example, suppose dx i =0.2, dy i =0.3, dy i-1 =0.25, dy i+1 = -0.3, obviously, dx i with dy i and dy i-1 If the signs are the same, that is, if dy is true, then dy is true. i and dy i-1 Two values ​​and dx i Since they have the same sign, we only need to compare dy. i and dy i-1 neutralization dx iDoes the largest difference exceed 1.5 times? (dy) i and dx i The difference is 0.1, dy i-1 and dx i The difference is 0.05, dy i and dx i The gap is greater than dy i-1 and dx i The gap is large, and Since the condition that the difference in the growth rate does not exceed the target multiple is met, I is 1.

[0062] It should be noted that, due to tc XY Due to the special nature of the formula construction, tc may exist. XY ≠tc YX Therefore, in this case, the mean can also be used as the final synchronous rate of change of feature X and feature Y, that is:

[0063] Before filtering features, invalid features can be removed first. That is, before calculating the correlation index between the feature data sequence of each feature and the feature data sequences of other features, the method in this embodiment further includes:

[0064] If the data missing rate of the feature data sequence of any feature among the m-dimensional features meets the first set condition, then delete that feature and its feature data.

[0065] Conversely, the degree of data homogeneity of the feature data sequence is determined.

[0066] If the degree of data homogeneity meets the second set condition, then the missing data in the feature data sequence of that feature is filled.

[0067] For example, if the missing data rate of a certain feature exceeds 30%, it indicates a high level of information loss in that feature, making it difficult to fill and correct through data imputation methods. Therefore, the feature and its data can be directly deleted. If the missing data rate does not exceed 30%, further analysis is needed. That is, after removing the missing feature data from the feature data sequence, observe whether the data homogeneity of the remaining samples for that feature is high. If the data homogeneity is high, the feature can also be deleted; otherwise, the missing data in the feature data sequence can be filled.

[0068] In one implementation, the degree of data homogeneity can be measured by the following factors: the volatility of the data distribution, whether the distribution is even, and whether the data clusters within a set range. For example, suppose there is a feature k, and its feature data sequence is represented as x. 1k x 2k, ..., x nk P samples are missing. Therefore, after removing the missing samples, the remaining samples are x. 1k x 2k , ..., x (n-p)k Then, we can judge this by measuring the degree of homogeneity of the remaining samples, as shown below:

[0069]

[0070] Where σ is x 1k x 2k , ..., x (n-p)k standard deviation It is its mean, q 0.95 and q 0.05 These are its 95% and 5% quantiles, m △ That is, for q 0.95 and q 0.05 The spacing between boxes is equal.

[0071] Based on s, m △ This allows analysis of whether the remaining sample data exhibits sufficiently high volatility, is evenly distributed, and whether data values ​​are clustered within a very small range. If the remaining sample data has very low volatility, or most of the data is concentrated within a small range, then the data quality is very low and should not be included in the model. For example, the following criteria can be constructed:

[0072] s>0.25

[0073]

[0074] Where P(q) 0.95 +i*m △ ) indicates that the sample falls within the interval [q 0.95 +(i-1)*m △ q 0.95 +i*m △ If all three of the above equations are satisfied, it can be considered that the homogeneity of the remaining sample data is not high, and this feature can be retained. Therefore, it can be introduced into the model after being filled and corrected by sample data.

[0075] The meaning and rationality of the above three equations will be explained next. 's' is essentially the coefficient of variation, used to measure the proportion of the standard deviation to the sample mean. Generally, a relatively dispersed and balanced data set follows the 4-sigma criterion, i.e., interval... The result roughly matches the 95% confidence interval of the sample. If s ≤ 0.25, then the 95% confidence interval of the sample is... The coverage area will be too small, and the fluctuation range of its upper and lower limits will be compressed to... Within this range, it is very detrimental to the discretization of data distribution and will seriously interfere with subsequent model fitting work. Secondly, This reflects q 0.95 and q 0.05 The maximum probability that each bin contains sample data after being divided into equally spaced bins. This is the minimum probability that the sample data is included after binning. Too large or Too small a probability indicates severe accumulation, meaning that the vast majority of data is concentrated in a small range, and some intervals have almost no samples, which means that the degree of homogeneity is too high. This is extremely bad for the balance of data distribution and does not benefit the fitting of subsequent models.

[0076] Therefore, the embodiments of this application can create a feature data homogeneity evaluation and quantification system by analyzing the missing rate of missing features, and perform preliminary feature screening based on this system to select data with sufficient information for subsequent modeling, thereby improving the quality of feature data.

[0077] Figure 2 This is a flowchart illustrating the process of determining features to be deleted in the feature filtering method provided in this application embodiment. This embodiment further optimizes the method for determining features to be deleted based on any of the above embodiments. Figure 2 As shown, the method includes:

[0078] S201. Obtain the correlation index of feature i and feature j with the feature data sequences of other features respectively, to obtain the correlation index sequence i and the correlation index sequence j.

[0079] S202. Based on the magnitude of each correlation index in correlation index sequence i and correlation index sequence j and the preset threshold, determine the features to be deleted from features i and features j.

[0080] Specifically, the magnitude of each correlation index in the correlation index sequence is used to measure which feature i and feature j has a greater overall correlation with all other features, and the feature with a greater overall correlation is selected as the feature to be deleted.

[0081] In one embodiment, the number of correlation indicators whose absolute values ​​exceed a preset threshold in correlation indicator sequence i and correlation indicator sequence j can be determined respectively, and the feature corresponding to the one with the larger number can be used as the feature to be deleted. That is, in correlation indicator sequences i and j, the one with the larger number of correlation indicators whose absolute values ​​exceed the preset threshold is used as the feature to be deleted.

[0082] In another embodiment, if the number of correlation indicators whose absolute values ​​exceed a preset threshold is the same in correlation indicator sequence i and correlation indicator sequence j, then the sum of the absolute values ​​of each correlation indicator in correlation indicator sequence i and correlation indicator sequence j is calculated respectively, and the feature corresponding to the larger sum is selected as the feature to be deleted. If the sums are still the same, then a feature is randomly selected as the feature to be deleted.

[0083] For example, firstly, regarding the Pearson correlation coefficient P, using the correlation coefficient table notation from above, the Pearson correlation coefficient table for each feature is denoted as follows:

[0084] 1 2 3 …… D 1 <![CDATA[pρ 11 ]]> <![CDATA[pρ 12 ]]> <![CDATA[pρ 13 ]]> …… <![CDATA[pρ 1D ]]> 2 <![CDATA[pρ 21 ]]> <![CDATA[pρ 22 ]]> <![CDATA[pρ 23 ]]> …… <![CDATA[pρ 2D ]]> 3 <![CDATA[pρ 31 ]]> <![CDATA[pρ 32 ]]> <![CDATA[pρ 33 ]]> …… <![CDATA[pρ 3D ]]> …… …… …… …… …… …… D <![CDATA[pρ D1 ]]> <![CDATA[pρ D2 ]]> <![CDATA[pρ D3 ]]> …… <![CDATA[pρ DD ]]>

[0085] In this table, the i-th row represents the Pearson correlation coefficient between the i-th feature and other features. The solution in this embodiment is to first select the two features with the highest absolute values ​​of their Pearson correlation coefficients, and then select one of these features as the feature to be deleted based on other relevant characteristics reflected by these two features. To avoid interference, the correlation coefficient {pρ} of each feature in the Pearson correlation coefficient table with itself needs to be calculated. ii The values ​​of ,i=1,…,D} are removed because their values ​​are 1, which has reached the upper limit of the correlation coefficient and will interfere with the screening of highly correlated features, so they are removed.

[0086] In one embodiment, assuming that the two features with the highest absolute values ​​of correlation coefficients in the Pearson correlation coefficient table are feature i and feature j, if the absolute value of their Pearson correlation coefficients does not exceed 0.5, it indicates that the correlation between all features is weak and they should all be retained. Conversely, if they exceed 0.5, then the correlation coefficients of each of feature i and feature j with other features are observed to obtain the Pearson correlation coefficient sequence h. i The correlation coefficient sequence h with Pearson j They are represented as follows:

[0087] h i =(pρ i1 , pρ i2 ..., pρ i(i-1) , pρ i(i+1) ..., pρ iD}

[0088] h j ={pρ j1 , pρ j2 ..., pρ j(j-1) , pρ j(j+1) ..., pρ jD}

[0089] h i with h jThe feature with the largest number of data points having a correlation coefficient absolute value exceeding 0.5 is selected as the feature to be deleted. For example, suppose h i Five of them had Pearson correlation coefficients with absolute values ​​exceeding 0.5, h j If seven of the correlation coefficients have absolute values ​​exceeding 0.5, it indicates that feature j has a higher overall correlation with other features, and the information from feature j is more easily expressed by other features; therefore, feature j should be deleted. If h i with h j If the number of correlation coefficients with absolute values ​​exceeding 0.5 is the same, then h is... i with h j The absolute values ​​of the Pearson correlation coefficients in the data are summed, and the feature with the higher sum is selected as the feature to be deleted. If the sums are still the same, a feature is randomly selected as the feature to be deleted.

[0090] The Pearson correlation coefficient table of the samples is reconstructed for the remaining features after deletion, and the same method as described in the embodiments of this application is used to determine whether there are any new features to be deleted. This process is iterated until no two features with an absolute correlation coefficient greater than 0.5 are found among the remaining features. This is equivalent to performing the first round of feature screening using the Pearson correlation coefficient.

[0091] For example, the features selected in the first round by Pearson correlation coefficient can be summarized and a Spearman correlation coefficient table can be constructed. Using the same logic and method as the Pearson correlation coefficient selection, the two features with the highest absolute value of Spearman correlation coefficient are selected and one of them is deleted until no two features with an absolute value of Spearman correlation coefficient exceeding 0.5 are found among all the remaining features. This is equivalent to performing a second round of feature selection using Spearman correlation coefficient.

[0092] Then, the features filtered by Spearman correlation coefficient are reorganized and summarized to construct a Kendall rank correlation coefficient table. Similarly, the two features with the highest absolute value of Kendall rank correlation coefficient are selected and one of them is removed until no two features with an absolute value of Kendall rank correlation coefficient exceeding 0.5 are found among all the remaining features. This is equivalent to performing a third round of feature filtering using Kendall rank correlation coefficient.

[0093] Finally, the features filtered by Kendall rank correlation coefficient are reorganized and summarized, and a synchronous rate of change table is constructed. Based on the synchronous rate of change table, feature selection is performed again according to the filtering logic above, which is equivalent to a fourth round of feature filtering using synchronous rate of change.

[0094] The final features obtained from the four rounds of screening are then organized and used for subsequent model fitting. Thus, by constructing a correlation analysis method and creating a synchronous rate of change analysis mechanism, a four-fold feature screening system is formed, achieving intelligent feature selection and greatly improving the effectiveness of feature screening.

[0095] Figure 3 This is a flowchart illustrating the feature correction method in the feature selection method provided in this application embodiment. This embodiment further optimizes the feature correction process based on any of the above embodiments. Here, the dimension of the target feature is represented as k. Figure 3 As shown, the method includes:

[0096] S301. Input the feature data of n samples under k-dimensional features into the target model for fitting.

[0097] S302. Calculate the evaluation index of the degree of influence of each of the k-dimensional features on the target variable of the target model.

[0098] S303. Randomly shuffle the numerical order of the target variable a certain number of times, and calculate the mean value of the evaluation index of the influence of each of the k dimensional features on the target variable after shuffling a certain number of times.

[0099] S304. Compare the influence evaluation indicators of each dimension before shuffling and the influence evaluation indicators after shuffling with the set reasonable range, and correct the k dimension features based on the comparison results.

[0100] The k-dimensional features obtained after screening are input into the model for fitting. During the fitting process, the model also provides feedback on the effectiveness of the features. This information can be used to further refine the features, removing unsuitable ones to achieve further feature simplification. This application does not limit the type of the target model; for example, it could be a predictive model used by financial institutions to predict the purchase intention of financial products.

[0101] Model fitting yields an assessment of the impact of each feature on the target variable of the target model, i.e., an impact evaluation index. Impact evaluation indices include at least one of the following: the IV value, which reflects the predictive power of the feature variable on the target variable; the importance value, which measures the magnitude of the impact of the feature variable on changes in the target variable; and the correlation coefficient between the feature and the target variable.

[0102] The IV value, or Information Value, is calculated by binning the range of values ​​for a feature, calculating the WOE (Weight of Evidence) value of each bin, and then weighting and summing them. The Importance value, or feature_importance, is a metric unique to tree models like LightGBM (Light Gradient Boosting Machine) that measures the influence of a feature variable on the target variable. It's important to note that for tree models, feature selection based on Importance doesn't rely on the specific Importance value between each feature and the target variable. Instead, it's determined by the ranking of feature i's Importance value among all features. If the model requires 20 features, the selection principle is based on whether its ranking is within the top 20. If it is, then feature i is marked as important. i =1, otherwise 0. The correlation coefficient between the feature and the target variable includes, but is not limited to, the Pearson correlation coefficient (p), the Spearman correlation coefficient (sp), and the Kendall rank correlation coefficient (tau). In one embodiment, to reflect the overall effect, the average of the three correlation coefficients can be taken as the final correlation coefficient ρ. * .

[0103] In one implementation, all indicators reflecting the relationship between features and the target variable are evaluated in a combined manner. Taking LightGBM as an example, this involves combining IV value, importance value, and ρ. * Feedback and adjustments are made based on a comprehensive consideration approach. Using the notation from above, let's assume the relationship between the k features and the target variable y is as shown in the table below:

[0104]

[0105]

[0106] Extract the feature data from each column in the table above and perform IV, importance, and ρ calculations with the target variable y. * The calculation, for example, the numerical value {x} of any column feature i. 1i ,x 2i ,…,x ni} and {y1,y2,…,y n The calculation result of} is labeled as IV. i importance i and Then the target variable {y1,y2,…,y nThe values ​​of} are shuffled, and after each shuffle, new IV, importance, and ρ can be calculated for feature i and the shuffled target variable y. * The value is determined by repeating this shuffling operation a target number of times, such as 1000 times. Each time, the different IVs, importance, and ρ values ​​related to feature i and the shuffled target variable y will be obtained. * Value. The average of the 1000 results is used as the mean of the evaluation index of the influence of feature i on the target variable y, denoted as . and

[0107] Next, for the impact evaluation indicators of each dimension feature before shuffling, if any feature's impact evaluation indicator is outside the set reasonable range, then that feature and its data are deleted. For any feature, if its impact evaluation indicator before shuffling is within the set reasonable range, but the mean of its impact evaluation indicator after shuffling is higher than its impact evaluation indicator before shuffling, then that feature and its data are deleted.

[0108] Therefore, for example, the following evaluation rule can be constructed:

[0109] (1) Start with the first feature (i=1) and perform a cyclical screening:

[0110] (2) Compare "IV" i importance i , "and The following criteria are used to determine whether feature i should be deleted:

[0111] When "IV" i >0.5” or “IV” i If < 0.1, discard feature i;

[0112] When "0.1≤IV" i ≤0.3” and When, discard feature i;

[0113] When "importance" i When = 0, discard feature i;

[0114] When "importance" i =1 and When, discard feature i;

[0115] when When, discard feature i;

[0116] when and When, discard feature i;

[0117] (3) After feature i is judged, start the next feature loop, i.e. i = i + 1, and repeat the loop judgment steps in (2);

[0118] (4) When i = k and the loop judgment step in (2) is completed, the entire loop filtering process ends;

[0119] The rationale for this approach is that if a feature is truly important to the model, then the IV, importance, and ρ corresponding to that feature should be considered first. * All should be within a reasonable range. For example, a reasonable range for IV is 0.1-0.5, and importance corresponds to the importance of the feature. i The reasonable value should be 1, ρ * The reasonable range should be at least greater than 0.1. However, merely being within a reasonable range is still insufficient. If a feature is truly important, then after the order of the target variable y is shuffled, it is equivalent to the original data being severely disturbed. In this case, the feature should change from an important feature to an unimportant feature. Therefore, further analysis and comparison are needed. If the IV value before the order is shuffled is still within 0.1 ≤ IV... i The reasonable range is ≤0.3, but disrupting it actually... If the IV value further increases, it means that shuffling the order of the target variable y not only did not reduce the importance of the features, but actually made the features seem more important, which clearly contradicts the facts. Therefore, the statement above, "when 0.1 ≤ IV", is incorrect. i ≤0.3 and The judgment rule is to discard feature i when it is not present. Similarly, the rule is to discard feature i when it is not present. i =1 and "Discard feature i" and "when" and The judgment rule for "discarding feature i" is based on the same logic.

[0120] The technical solution of this application embodiment creates a model fitting feedback correction mechanism to fit the selected features to the model, and corrects these features based on the model fitting effect. By analyzing the changes in feature importance after shuffling the order of target variables, the rationality of feature selection is further improved, laying the foundation for real-time updates of model features and greatly reducing the threshold for beginners to use model technology.

[0121] Figure 4 This is a schematic diagram of the feature screening device provided in an embodiment of this application. Figure 4 As shown, the feature screening device 40 includes:

[0122] The feature data acquisition module 410 is used to acquire feature data tables of n samples under m-dimensional features, and to take the n feature data under any dimension feature as a feature data sequence.

[0123] The correlation index calculation module 420 is used to calculate the correlation index between the feature data sequence of each feature and the feature data sequence of other features, and obtain the correlation index table.

[0124] The selection module 430 is used to select feature i and feature j corresponding to the correlation index with the highest absolute value from the correlation index table;

[0125] The feature to be deleted determination module 440 is used to determine the feature to be deleted from feature i and feature j based on the correlation between feature i and feature j and other features if the correlation index with the highest absolute value exceeds a preset threshold.

[0126] Feature deletion module 450 is used to delete the feature to be deleted and its feature data sequence to obtain a new feature data table;

[0127] The loop processing module 460 is used to return to the step of calculating the correlation index between the feature data sequence of each feature and the feature data sequence of other features for the new feature data table, until the correlation index with the highest absolute value in the correlation index table of the new feature data table does not exceed the preset threshold, and the remaining features at this time are used as the target features obtained by filtering.

[0128] In some embodiments, the feature-to-be-deleted determination module 440 includes:

[0129] The correlation index sequence acquisition unit is used to acquire the correlation index of feature i and feature j with other feature data sequences respectively, so as to obtain correlation index sequence i and correlation index sequence j.

[0130] The feature to be deleted determination unit is used to determine the feature to be deleted from feature i and feature j based on the magnitude of each correlation index in the correlation index sequence i and the correlation index sequence j and the preset threshold.

[0131] In some embodiments, the feature-to-be-deleted determination unit is specifically used for:

[0132] The number of correlation indices whose absolute values ​​exceed the preset threshold in the correlation index sequence i and the correlation index sequence j are determined respectively, and the feature corresponding to the larger number is taken as the feature to be deleted.

[0133] In some embodiments, the feature-to-be-deleted determination unit is further configured to:

[0134] If the number of correlation indicators whose absolute values ​​exceed the preset threshold is the same in the correlation indicator sequence i and the correlation indicator sequence j, then the sum of the absolute values ​​of each correlation indicator in the correlation indicator sequence i and the correlation indicator sequence j is calculated respectively, and the feature corresponding to the larger sum is taken as the feature to be deleted.

[0135] In some embodiments, the correlation index includes at least one of the following: Pearson correlation coefficient, Spearman correlation coefficient, Kendall rank correlation coefficient, and synchronous rate of change, wherein the synchronous rate of change is used to observe the phenomenon of synchronous changes in data over a certain time span.

[0136] In some embodiments, the synchronous rate of change is expressed by the following formula:

[0137]

[0138] dx i =(x i -x i-1 ) / x i-1 i = 2, ..., n

[0139] dy i =(y i -y i-1 ) / y i-1 i = 2, ..., n

[0140] Where X and Y represent any two features, tc XY dx represents the synchronous rate of change of feature X and feature Y. i and dy i Let x represent the data in the feature data sequence of feature X and feature Y respectively. i Data y i The percentage increase;

[0141] For synchronous characteristic functions, when dx i with dy i dy i-1 dy i+1 If any two positive and negative numbers have the same sign, and their growth rates differ by no more than the target multiple, then I is 1; otherwise, I is 0.

[0142] In some embodiments, the dimension number of the target feature is represented as k, and the apparatus further includes a correction module, the correction module comprising:

[0143] The fitting unit is used to input the feature data of the n samples under k-dimensional features into the target model for fitting.

[0144] The influence degree evaluation index calculation unit is used to calculate the influence degree evaluation index of each of the k-dimensional features on the target variable of the target model.

[0145] The sequence shuffling unit is used to randomly shuffle the numerical order of the target variable a certain number of times, and calculate the average value of the evaluation index of the influence of each of the k dimensional features on the target variable after shuffling a certain number of times.

[0146] The correction unit is used to compare the influence evaluation index of each dimension feature before shuffling, the mean of the influence evaluation index after shuffling, and the relationship with a set reasonable range, and to correct the k dimension features based on the comparison results.

[0147] In some embodiments, the correction unit is specifically used for:

[0148] Regarding the evaluation indicators for the degree of influence of each dimension's features before shuffling, if any feature's degree of influence evaluation indicator falls outside the set reasonable range, then that feature and its data will be deleted; and

[0149] For any feature, if its impact index before scrambling is within the set reasonable range, but the average impact index after scrambling is higher than the impact index before scrambling, then delete the feature and its data.

[0150] In some embodiments, the impact assessment index includes at least one of the following:

[0151] The IV value is used to reflect the predictive power of a feature variable on a target variable; the importance value is used to measure the magnitude of the influence of a feature variable on changes in the target variable; and the correlation coefficient between the feature and the target variable is used.

[0152] In some embodiments, the apparatus further includes a data preprocessing module, the data preprocessing module being specifically used for:

[0153] Before the correlation index calculation module 420 calculates the correlation index between the feature data sequence of each feature and the feature data sequence of other features, if the data missing rate of the feature data sequence of any feature among the m-dimensional features meets the first set condition, then the feature and its feature data are deleted.

[0154] Conversely, the degree of data homogeneity of the feature data sequence is determined.

[0155] If the degree of data homogeneity meets the second set condition, then the missing data in the feature data sequence of that feature is filled.

[0156] In some embodiments, the degree of data homogeneity is determined by at least one of the following methods:

[0157] The magnitude of data distribution volatility, whether the distribution is even, and whether the data clusters within a set range.

[0158] In some embodiments, the features obtained after the correction are used to train the target model.

[0159] The feature filtering device provided in this application embodiment can be used to execute the technical solution of the feature filtering method in the above embodiment. Its implementation principle and technical effect are similar, and will not be described again here.

[0160] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented in software via processing element calls, while others are implemented in hardware. For example, the correlation index calculation module 420 can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its function can be called and executed by a processing element of the above device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through the integrated logic circuits in the hardware of the processor element or through software instructions.

[0161] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device may include: transceiver 121, processor 122, and memory 123.

[0162] Processor 122 executes computer execution instructions stored in memory, causing processor 122 to perform the scheme in the above embodiments. Processor 122 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0163] The memory 123 is connected to the processor 122 via the system bus and completes communication between them. The memory 123 is used to store computer program instructions.

[0164] Transceiver 121 can be used to obtain the task to be run and its configuration information.

[0165] The system bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.

[0166] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.

[0167] This application also provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the feature filtering method described in the above embodiments.

[0168] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium, and when the at least one processor executes the computer program, it can implement the technical solution of the feature screening method in the above embodiments.

[0169] In the implementation of the computer program product, computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0170] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A feature selection method, characterized in that, include: Obtain a feature data table of n samples under m dimensions, and take the n feature data under any dimension as a feature data sequence; The correlation index between the feature data sequence of each feature and the feature data sequence of other features is calculated to obtain a correlation index table. The correlation index includes the synchronous change rate, which is used to observe the phenomenon of the same frequency of data change within a certain time span. The synchronous rate of change is expressed by the following formula: Where X and Y represent any two features, This represents the synchronous rate of change of feature X and feature Y. and These represent the data in the feature data sequences of feature X and feature Y, respectively. ,data The growth rate The first feature in the feature data sequence representing feature X. The original data of each sample; The first feature in the feature data sequence representing feature Y The original data of each sample; For synchronous indicator functions, when and If any two positive and negative values ​​have the same sign, and their growth rates differ by no more than the target multiple, then I is 1; otherwise, I is 0. Select feature i and feature j corresponding to the correlation index with the highest absolute value from the correlation index table; If the correlation index with the highest absolute value exceeds the preset threshold, then the feature to be deleted is determined from feature i and feature j based on the correlation between feature i and feature j and other features. The feature to be deleted and its feature data sequence are deleted to obtain a new feature data table; For the new feature data table, return to the step of calculating the correlation index between the feature data sequence of each feature and the feature data sequence of other features, until the correlation index with the highest absolute value in the correlation index table of the new feature data table does not exceed the preset threshold, and use the remaining features at this time as the target features obtained by filtering.

2. The method according to claim 1, characterized in that, The step of determining the feature to be deleted from feature i and feature j based on the correlation between each of feature i and feature j and all other features includes: Obtain the correlation index of feature i and feature j with other feature data sequences respectively, to obtain correlation index sequence i and correlation index sequence j; Based on the magnitude of each correlation index in the correlation index sequence i and the correlation index sequence j and the preset threshold, the feature to be deleted is determined from the feature i and feature j.

3. The method according to claim 2, characterized in that, The step of determining the feature to be deleted from feature i and feature j based on the magnitude of each correlation index in the correlation index sequence i and the correlation index sequence j and the preset threshold includes: The number of correlation indices whose absolute values ​​exceed the preset threshold in the correlation index sequence i and the correlation index sequence j are determined respectively, and the feature corresponding to the larger number is taken as the feature to be deleted.

4. The method according to claim 3, characterized in that, The step of determining the feature to be deleted from feature i and feature j based on the magnitude of each correlation index in the correlation index sequence i and the correlation index sequence j and the preset threshold further includes: If the number of correlation indicators whose absolute values ​​exceed the preset threshold is the same in the correlation indicator sequence i and the correlation indicator sequence j, then the sum of the absolute values ​​of each correlation indicator in the correlation indicator sequence i and the correlation indicator sequence j is calculated respectively, and the feature corresponding to the larger sum is taken as the feature to be deleted.

5. The method according to any one of claims 1-4, characterized in that, The correlation index includes at least one of the following: Pearson correlation coefficient, Spearman correlation coefficient, Kendall rank correlation coefficient, and synchronous rate of change.

6. The method according to claim 1, characterized in that, The dimension of the target feature is represented as k, and the method further includes: The feature data of the n samples under k-dimensional features are input into the target model for fitting; Calculate the evaluation index of the degree of influence of each of the k-dimensional features on the target variable of the target model; The numerical order of the target variable is randomly shuffled a number of times, and the mean value of the evaluation index of the influence of each of the k dimensional features on the target variable after shuffling is calculated respectively. The influence evaluation indicators of each dimension feature before shuffling and the mean of the influence evaluation indicators after shuffling are compared with the set reasonable range, and the k dimension features are corrected according to the comparison results.

7. The method according to claim 6, characterized in that, The comparison involves evaluating the impact of each dimension feature before scrambling, the mean of the impact evaluation indicators after scrambling, and their relationship to a set reasonable range. Based on the comparison results, the k dimension features are then corrected, including: Regarding the evaluation indicators for the degree of influence of each dimension's features before shuffling, if any feature's degree of influence evaluation indicator falls outside the set reasonable range, then that feature and its data will be deleted; and For any feature, if its impact index before scrambling is within the set reasonable range, but the average impact index after scrambling is higher than the impact index before scrambling, then delete the feature and its data.

8. The method according to claim 6 or 7, characterized in that, The evaluation indicators for the degree of impact include at least one of the following: The IV value is used to reflect the predictive power of a feature variable on a target variable; the importance value is used to measure the magnitude of the influence of a feature variable on changes in the target variable; and the correlation coefficient between the feature and the target variable is used.

9. The method according to claim 1, characterized in that, Before calculating the correlation index between the feature data sequence of each feature and the feature data sequences of other features, the method further includes: If the data missing rate of the feature data sequence of any feature among the m-dimensional features meets the first set condition, then delete that feature and its feature data. Conversely, the degree of data homogeneity of the feature data sequence is determined. If the degree of data homogeneity meets the second set condition, then the missing data in the feature data sequence of that feature is filled.

10. The method according to claim 9, characterized in that, The degree of data homogeneity is determined by at least one of the following methods: The magnitude of data distribution volatility, whether the distribution is even, and whether the data clusters within a set range.

11. The method according to claim 6, characterized in that, The features obtained after the correction are used to train the target model.

12. A feature screening device, characterized in that, include: The feature data acquisition module is used to acquire feature data tables of n samples under m-dimensional features, and to take the n feature data under any dimension as a feature data sequence. The correlation index calculation module is used to calculate the correlation index between the feature data sequence of each feature and the feature data sequence of other features to obtain a correlation index table. The correlation index includes the synchronous change rate, which is used to observe the phenomenon of the same frequency of data change within a certain time span. The synchronous rate of change is expressed by the following formula: Where X and Y represent any two features, This represents the synchronous rate of change of feature X and feature Y. and These represent the data in the feature data sequences of feature X and feature Y, respectively. ,data The growth rate The first feature in the feature data sequence representing feature X. The original data of each sample; The first feature in the feature data sequence representing feature Y The original data of each sample; For synchronous indicator functions, when and If any two positive and negative values ​​have the same sign, and their growth rates differ by no more than the target multiple, then I is 1; otherwise, I is 0. The selection module is used to select feature i and feature j corresponding to the correlation index with the highest absolute value from the correlation index table; The feature to be deleted module is used to determine the feature to be deleted from feature i and feature j based on the correlation between feature i and feature j and other features if the correlation index with the highest absolute value exceeds a preset threshold. The feature deletion module is used to delete the feature to be deleted and its feature data sequence to obtain a new feature data table; The loop processing module is used to return to the step of calculating the correlation index between the feature data sequence of each feature and the feature data sequence of other features for the new feature data table, until the correlation index with the highest absolute value in the correlation index table of the new feature data table does not exceed the preset threshold, and the remaining features at this time are used as the target features obtained by filtering.

13. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the feature screening method as described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the feature selection method as described in any one of claims 1-11.

15. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the feature selection method according to any one of claims 1-11.

Images