Feature screening method, early warning method, device, electronic device, medium and program

Abstract

The present disclosure provides a feature screening method, an early warning method, an apparatus, an electronic device, a medium and a program. The feature screening method, the early warning method and the apparatus can be used in the field of financial technology. The feature screening method for data includes: determining a data sample, the data sample includes multiple features; screening the data in the first stage; and screening the data in the second stage. Among them, the first stage of screening the data includes: using the maximum correlation-minimum redundancy algorithm, the feature weight algorithm, and the T-test algorithm to perform feature screening on the data sample respectively, and obtaining the first stage feature subset by merging the screening results. Among them, the second stage of screening the data includes: using the packaged feature selection method and the embedded feature selection method to obtain the final feature.

Description

Technical Field

[0001] The present disclosure relates to the field of financial technology, and more specifically, to a method for feature screening of data, an early warning method for bank customer loss, a device, an electronic device, a medium and a computer program. Background Art

[0002] The development of Internet finance has brought many new challenges to traditional banks. For traditional banks, how to retain existing customers in an increasingly sparse customer market is particularly important, and how to identify potential lost customers in advance and retain them is the focus of banks. At present, in order to address the problem of bank customer loss warning, the industry mostly uses limited customer data to classify customer loss status. However, with the development of the big data era, banks have accumulated a large amount of high-dimensional customer data. The traditional model has achieved limited results, and the efficiency of the classification algorithm and the accuracy of the classification need to be improved.

[0003] Therefore, how to use the effective information in high-dimensional customer data to build an early warning model to determine whether customers are likely to churn in the future, and then help decision makers discover problems early and adopt corresponding retention strategies to reduce customer churn and reduce bank losses is an issue that needs to be urgently addressed in the industry. Summary of the invention

[0004] In view of this, the present disclosure provides a feature screening method for data, an early warning method for bank customer churn, a device, an electronic device, a computer-readable storage medium and a computer program with a simple algorithm, low cost and high classification accuracy.

[0005] One aspect of the present disclosure provides a method for feature screening of data, including the following operations: determining a data sample, wherein the data sample includes a plurality of features. The first stage of screening data includes: using a maximum correlation-minimum redundancy algorithm, using a feature weight algorithm, and using a T-test algorithm to perform feature screening on the data sample respectively, and obtaining a first stage feature subset by merging the screening results; the second stage of screening data includes: using a packaged feature selection method to perform feature screening on the first stage feature subset to obtain a fourth feature subset; using an embedded feature selection method to perform feature screening on the first stage feature subset to obtain a fifth feature subset; and obtaining the final feature based on the fourth feature subset and the fifth feature subset.

[0006] According to the feature screening method for data in the embodiment of the present disclosure, by screening data in the first stage, the maximum relevance-minimum redundancy algorithm, feature weight algorithm and T-test algorithm are used to reduce the dimension of the original features, and then select the features that have a greater impact on judging whether bank customers have churned, that is, to obtain the first-stage feature subset. The first-stage feature subset is used as the input feature for the second-stage screening of data, thereby further reducing the complexity of the classification algorithm, reducing the computational cost, and improving the classification accuracy.

[0007] In some embodiments, the use of the packaged feature selection method to perform feature screening on the first-stage feature subset includes iterating the first-stage feature subset using a feature ranking algorithm, and obtaining a fourth feature subset during the iteration process.

[0008] In some embodiments, the use of the embedded feature selection method to perform feature screening on the first-stage feature subset includes iterating the first-stage feature subset using a gradient enhancement algorithm, and obtaining a fifth feature subset during the iteration process.

[0009] In some embodiments, obtaining the final feature according to the fourth feature subset and the fifth feature subset includes calculating the importance score of each feature according to the indicator function, arranging the importance scores in ascending or descending order, and obtaining N final The final features, N final is an integer greater than or equal to 1, and the indicator function includes the following formula:

[0010]

[0011]

[0012] Among them, f i represents each feature; m represents the model of the packaged feature selection method or the model of the embedded feature selection method; ti represents the number of the fourth feature subset or the fifth feature subset; acc m，k represents the accuracy of model m at the kth run; N final Represents the number of final features; r(f i , m, k) represents the feature f of model m at the kth run i The maximum rank of the features in each model m is recorded as

[0013] In some embodiments, determining the data sample includes: acquiring first data from a data storage system, and using the first data as the data sample.

[0014] In some embodiments, determining the data sample includes: acquiring first data from a data storage system; and regressing abnormal feature values ​​in the first data to obtain second data, and using the second data as the data sample.

[0015] In some embodiments, the abnormal feature value includes a missing value, and regressing the abnormal feature value in the first data includes filling the missing value with the mean or median of the data under the abnormal feature, or determining the missing value by induction.

[0016] In some embodiments, determining the data sample includes: acquiring first data from the plurality of data storage systems respectively; regressing outliers in the first data to obtain second data; and integrating the second data obtained from different data storage systems to form the data sample.

[0017] In some embodiments, the determining of the data sample further comprises: performing a standardized conversion on the integrated second data using a standardized method, and the data after the standardized conversion is the data sample.

[0018] Another aspect of the present disclosure provides an early warning method for bank customer churn, comprising: executing the feature screening method for data as described above to obtain final features; inputting the final features into a model as an input layer; converting the feature data input into the input layer into intermediate data; and converting the intermediate data into a final result, wherein the final result is used to characterize the customer as a customer prone to churn or a customer not prone to churn.

[0019] According to the bank customer loss early warning method of the embodiment of the present disclosure, it is possible to introduce the feature screening method for data into the bank customer loss early warning method, and then it is possible to predict which of the bank's customers are stable customers and which are customers that are about to be lost, so that the bank can retain the customers that are about to be lost in a targeted manner and improve the bank's service level and performance.

[0020] In some embodiments, converting the intermediate data into the final result includes: applying a sigmoid function, wherein the sigmoid function calculates a mapping value between (0, 1); and comparing the mapping value with a standard threshold to obtain the final result.

[0021] Another aspect of the present disclosure provides an apparatus including a data determination module, a first screening module, and a second screening module.

[0022] The data determination module is used to determine the data samples and determine the data samples that need to be screened, and the data samples include multiple features; the first screening module is used to perform the first stage of data screening, and the first stage of data screening includes: using the maximum relevance-minimum redundancy algorithm, using the feature weight algorithm and using the T-test algorithm to perform feature screening on the data samples respectively, and the screening results are combined to obtain the first stage feature subset.

[0023] The second screening module is used to perform second-stage screening of data, and the second-stage screening of data includes: using a packaged feature selection method to perform feature screening on the first-stage feature subset to obtain a fourth feature subset; using an embedded feature selection method to perform feature screening on the first-stage feature subset to obtain a fifth feature subset; and obtaining the final features based on the fourth feature subset and the fifth feature subset.

[0024] Another aspect of the present disclosure provides an electronic device, comprising one or more processors and one or more memories, wherein the memories are used to store executable instructions, and when the executable instructions are executed by the processors, the above-mentioned method is implemented.

[0025] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the above method when executed.

[0026] Another aspect of the present disclosure provides a computer program, which includes computer executable instructions, and the instructions are used to implement the method described above when executed. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0028] Figure 1 An exemplary system architecture to which the method and apparatus according to an embodiment of the present disclosure can be applied is schematically shown;

[0029] Figure 2 A flowchart of a method for feature screening of data according to an embodiment of the present disclosure is schematically shown;

[0030] Figure 3 Schematically shows a flowchart of the first stage of screening data according to an embodiment of the present disclosure;

[0031] Figure 4 Schematically shows a flow chart of the second stage of screening data according to an embodiment of the present disclosure;

[0032] Figure 5A flowchart of using a packaged feature selection method to perform feature screening on a first-stage feature subset according to an embodiment of the present disclosure is schematically shown;

[0033] Figure 6 A flowchart of using an embedded feature selection method to perform feature screening on a first-stage feature subset according to an embodiment of the present disclosure is schematically shown;

[0034] Figure 7 A flowchart for determining a data sample according to an embodiment of the present disclosure is schematically shown;

[0035] Figure 8 A flowchart for determining a data sample according to an embodiment of the present disclosure is schematically shown;

[0036] Fig. 9 A flowchart for determining a data sample according to an embodiment of the present disclosure is schematically shown;

[0037] Fig.10 A flowchart schematically shows a method for early warning of bank customer loss according to an embodiment of the present disclosure;

[0038] Fig.11 A flowchart of converting intermediate data into final results according to an embodiment of the present disclosure is schematically shown;

[0039] Fig.12 A block diagram schematically shows a device according to an embodiment of the present disclosure;

[0040] Fig.13 A block diagram of an electronic device according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0041] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure. In the technical solution of the present disclosure, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations, necessary confidentiality measures are taken, and do not violate public order and good morals.

[0042] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0043] When using expressions such as "at least one of A, B or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of the said features.

[0044] The development of Internet finance has brought many new challenges to traditional banks. For traditional banks, how to retain existing customers in an increasingly sparse customer market is particularly important, and how to identify potential lost customers in advance and retain them is the focus of banks. At present, in order to address the problem of bank customer loss warning, the industry mostly uses limited customer data to classify customer loss status. However, with the development of the big data era, banks have accumulated a large amount of high-dimensional customer data. The traditional model has achieved limited results, and the efficiency of the classification algorithm and the accuracy of the classification need to be improved.

[0045] Therefore, how to use the effective information in high-dimensional customer data to build an early warning model to determine whether customers are likely to churn in the future, and then help decision makers discover problems early and adopt corresponding retention strategies to reduce customer churn and reduce bank losses is an issue that needs to be urgently addressed in the industry.

[0046] The embodiments of the present disclosure provide a method for feature screening of data, a method for early warning of bank customer loss, a device, an electronic device, a computer-readable storage medium, and a computer program. The method for feature screening of data includes: determining a data sample, the data sample includes multiple features; screening data in the first stage; and screening data in the second stage.

[0047] The first stage of data screening includes: using the maximum relevance-minimum redundancy algorithm, the feature weight algorithm and the T-test algorithm to perform feature screening on the data samples respectively, and obtaining the first stage feature subset by merging the screening results.

[0048] Among them, the second stage of data screening includes: using the packaged feature selection method to perform feature screening on the first stage feature subset to obtain the fourth feature subset; using the embedded feature selection method to perform feature screening on the first stage feature subset to obtain the fifth feature subset; and obtaining the final feature by taking the union of the fourth feature subset and the fifth feature subset.

[0049] The early warning method for bank customer churn includes: executing the feature screening method for data as described above to obtain final features; inputting the final features into a model as an input layer; converting the feature data input into the input layer into intermediate data; and converting the intermediate data into a final result, wherein the final result is used to characterize a customer as a customer prone to churn or a customer not prone to churn.

[0050] It should be noted that the feature screening method for data, the early warning method for bank customer churn, the device, the electronic device, the computer-readable storage medium and the computer program disclosed in the present invention can be used in the financial field, and can also be used in any field other than the financial field. The field of the present invention is not limited here.

[0051] Figure 1 The exemplary system architecture 100 of the method for feature screening of applied data, the method for early warning of bank customer churn, the device, the electronic device, the computer-readable storage medium and the computer program according to the embodiment of the present disclosure is schematically shown. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0052] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0053] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

[0054] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0055] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0056] It should be noted that the feature screening method for data and the early warning method for bank customer loss provided in the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the device provided in the embodiments of the present disclosure can generally be set in the server 105. The feature screening method for data and the early warning method for bank customer loss provided in the embodiments of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the device provided in the embodiments of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0057] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0058] The following will be based on Figure 1 The scene described by Figure 2 to Figure 9 The feature screening method for data of the disclosed embodiment is described in detail. The feature screening method for data can perform feature screening on any data that requires feature screening. The following is only described by taking the feature screening to determine whether a bank's customers are about to churn as an example, but it should not be understood as a limitation of the present disclosure. The present disclosure can also be used for, but not limited to, biometric screening, etc.

[0059] Figure 2 The flowchart of the method for feature screening of data according to an embodiment of the present disclosure is schematically shown.

[0060] like Figure 2 As shown, the feature screening method of this embodiment includes operations S210 to S230.

[0061] In operation S210, a data sample is determined, and the data sample includes multiple features. For example, the data sample may be account opening information of each customer of the bank, and the multiple features may be understood as various items in the account opening information. For example, the bank has customers Zhang San, Li Si, Wang Wu, Danny, and Li Ming, etc., and the account opening information of each customer includes name, age, work unit, home address, telephone number, monthly income, and balance, etc.

[0062] Therefore, data sample 1 is the account opening information of customer Zhang San, including feature 1: name; feature 2: age; feature 3: work unit; feature 4: home address; feature 5: telephone number; feature 6: monthly deposit; feature 7: balance, etc.

[0063] Data sample 2 is the account opening information of customer Li Si, including feature 1: name; feature 2: age; feature 3: work unit; feature 4: home address; feature 5: telephone number; feature 6: monthly deposits; feature 7: balance, etc.

[0064] Data sample 3 is the account opening information of customer Wang Wu, including feature 1: name; feature 2: age; feature 3: work unit; feature 4: home address; feature 5: telephone number; feature 6: monthly deposits; feature 7: balance, etc.

[0065] Data sample 4 is the account opening information of customer Danny, including feature 1: name; feature 2: age; feature 3: work unit; feature 4: home address; feature 5: telephone number; feature 6: monthly deposits; feature 7: balance, etc.

[0066] Data sample 5 is the account opening information of customer Li Ming, including feature 1: name; feature 2: age; feature 3: work unit; feature 4: home address; feature 5: telephone number; feature 6: monthly deposit; feature 7: balance, etc.

[0067] In operation S220, the first stage screens the data, such as Figure 3 As shown, the first stage of data screening in operation S220 includes operations S221 to S224.

[0068] In operation S221, a maximum relevance-minimum redundancy algorithm (mRMR algorithm) is used to perform feature screening on the determined data samples to obtain a first feature subset.

[0069] It should be noted that the maximum relevance-minimum redundancy algorithm includes formula (1).

[0070]

[0071] Among them, S represents the feature set, that is, the set of features in the account opening information; c represents the category of the data sample, which can be understood as the category of the customer, such as easy-to-churn customers or non-easy-to-churn customers; f i and f j Represents each feature, I(f i , c) is the mutual information between each feature and the category of the data sample, I(f i , f j ) is the mutual information between each feature.

[0072] The mutual information I(X; Y) can be obtained by formula (2).

[0073]

[0074] Among them, p(X;Y) represents the probability of X and Y appearing at the same time; p(X) represents the probability of X appearing alone; and p(Y) represents the probability of Y appearing alone. i Replace X, f j Substituting Y into formula (2) yields I(f i , f j ) value, and f i Substituting X and c for Y into formula (2) yields I(f i , c) value.

[0075] Therefore, the mRMR value of each feature can be obtained by formula (1) and formula (2). As an implementable method, the mRMR value of each feature can be compared with a set first threshold value, and the features with mRMR values ​​greater than the first threshold value are placed in the first feature subset, and the features with mRMR values ​​less than the first threshold value are eliminated. As another implementable method, the mRMR values ​​of all features can be sorted, and then the first m features are taken and placed in the first feature subset, and the remaining features are eliminated, where m can be an integer greater than or equal to 1.

[0076] In operation S222, a feature weight algorithm (Relief algorithm) is used to perform feature screening on the determined data sample to obtain a second feature subset.

[0077] It should be noted that the feature weight algorithm includes formula (3).

[0078] W(i)=W(i+1)-diff(A, R, H) / m+diff(A, R, M) / m (3)

[0079] Where W(i) is the feature weight value of the ith feature; W(i+1) is the feature weight value of the i+1th feature; A represents each feature; R represents the set of features of a certain category of the data sample; H represents a set of n features randomly selected from the set of the same category as R, where n is less than or equal to the number of features in the set of the same category as R; M represents a set of t features randomly selected from the set of different categories from R, where t is less than or equal to the number of features in the set of different categories from R; diff(A, R, H) represents the distance formula, that is, the correlation between R and H under feature A; diff(A, R, M) represents the correlation between R and M under feature A; m represents the number of samplings. Each time sampling, a sample R will be randomly selected from the data sample, and the correlation between R and H and the correlation between R and M under the ith feature will be calculated respectively. In other words, each feature A will be substituted into formula (3) and repeated m times, and each feature A will obtain the average weight after m repetitions. The initial weight of each feature A can be set to 0.

[0080] The distance formula includes discrete distance formula (4) and continuous distance formula (5). Discrete distance formula (4) is used for qualitative features, such as name, gender, etc.; continuous distance formula (5) is used for quantitative features, such as age, balance, etc.

[0081]

[0082]

[0083] Among them, value(A, I1) represents the value of sample I1 on feature A; value(A, I2) represents the value of sample I2 on feature A.

[0084] Therefore, the feature weight value of each feature can be obtained by formula (3) to formula (5). As an implementable method, the feature weight value of each feature can be compared with the set second threshold value, and the features with feature weight values ​​greater than the second threshold value are placed in the second feature subset, and the features with feature weight values ​​less than the second threshold value are eliminated. As another implementable method, the feature weight values ​​of all features can be sorted, and then the first h features are taken and placed in the second feature subset, and the remaining features are eliminated, where h can be an integer greater than or equal to 1.

[0085] In operation S223, a T-test algorithm is used to perform feature screening on the determined data sample to obtain a third feature subset.

[0086] It should be noted that the T-test algorithm includes formula (6).

[0087]

[0088] in, Represents the mean of each feature X1; Represents the mean of category X2 of the data sample; n1 is the sample size of feature X1; n2 is the sample size of category X2; S1 is the variance of feature X1 under all data samples; S2 is the variance of category X2 under all data samples.

[0089] Therefore, the T value of each feature can be obtained by formula (6). As an implementable method, the T value of each feature can be compared with the set third threshold, and the features with T values ​​greater than the third threshold are placed in the third feature subset, and the features with T values ​​less than the third threshold are eliminated. As another implementable method, the T values ​​of all features can be sorted, and then the first r features are taken and placed in the third feature subset, and the remaining features are eliminated, where r can be an integer greater than or equal to 1.

[0090] In operation S224 , a first-stage feature subset is obtained by obtaining a union of the first feature subset, the second feature subset, and the third feature subset.

[0091] In operation S230, the second stage screens the data, such as Figure 4 As shown, the second stage of data screening in operation S230 includes operations S231 to S233.

[0092] In operation S231, the first-stage feature subset is subjected to feature screening using a packaged feature selection method to obtain a fourth feature subset. It is understandable that the packaged feature selection method can be used to train samples containing the first-stage feature subset in each iteration, and the score ranking of all features can be obtained in each iteration process, and the feature scores are sorted from large to small, and the feature with the smallest score is removed, and then the next iteration is performed, and the sample model is trained again with the remaining features. Alternatively, a threshold is set, or the required number of features is set, and the fourth feature subset is selected.

[0093] In operation S232, the first-stage feature subset is subjected to feature screening using an embedded feature selection method to obtain a fifth feature subset. It is understandable that the embedded feature selection method can be used to train samples containing the first-stage feature subset in each iteration, and the score ranking of all features can be obtained in each iteration process, and the feature scores are sorted from large to small, and the feature with the smallest score is removed, and then the next iteration is performed, and the sample model is trained again with the remaining features. Alternatively, a threshold is set, or the required number of features is set, and the fifth feature subset is selected.

[0094] In operation S233, the final feature is obtained according to the fourth feature subset and the fifth feature subset. As a possible implementation method, the final feature is obtained according to the fourth feature subset and the fifth feature subset, including calculating the importance score of each feature according to the indicator function, arranging the importance scores in ascending or descending order, and obtaining N final The final features, N final is an integer greater than or equal to 1, and the indicator function includes formula (7) and formula (8).

[0095]

[0096]

[0097] Among them, f i represents each feature; m represents the model of the packaged feature selection method or the model of the embedded feature selection method; t i represents the number of the fourth feature subset or the fifth feature subset; acc m，k represents the accuracy of model m at the kth run; N final represents the number of final features, that is, the set of features selected from the account opening information; r(f i , m, k) represents the feature f of model m at the kth run i The maximum rank of the features in each model m is recorded as

[0098] According to the feature screening method for data in the embodiment of the present disclosure, by screening data in the first stage, the maximum relevance-minimum redundancy algorithm, feature weight algorithm and T-test algorithm are used to reduce the dimension of the original features, and then select the features that have a greater impact on judging whether bank customers have churned, that is, to obtain the first-stage feature subset. The first-stage feature subset is used as the input feature for the second-stage screening of data, thereby further reducing the complexity of the classification algorithm, reducing the computational cost, and improving the classification accuracy.

[0099] The present disclosure also obtains the ordering of features, that is, rank, by combining the packaged feature selection method and the embedded feature selection method, and further obtains a method for calculating the feature importance value by combining the indicator function to obtain the final feature. The calculation method of the indicator function can balance the result deviation caused by different feature selection methods.

[0100] Figure 5 The flowchart of using the packaged feature selection method to perform feature screening on the first-stage feature subset according to an embodiment of the present disclosure is schematically shown.

[0101] Operation S231 uses a packaged feature selection method to perform feature screening on the first-stage feature subset, including operation S2311.

[0102] In operation S2311, a feature ranking algorithm (SVM-RFE algorithm) is used to iterate the first-stage feature subset, and a fourth feature subset is obtained during the iteration process.

[0103] It should be noted that the feature ranking algorithm includes formula (9) and formula (10).

[0104] C i =ω i 2 (9)

[0105]

[0106] Among them, C i represents the feature score of the i-th feature; ω i represents the weight of feature i; α i is a constant; i represents the category to which the i-th data sample belongs; x i Represents all the feature values ​​of the i-th data sample.

[0107] Therefore, the feature score of each feature can be obtained by formula (9) and formula (10). As an implementable method, the feature score of each feature can be compared with the set fourth threshold, and the features with feature scores greater than the third threshold are placed in the fourth feature subset, and the features with feature scores less than the fourth threshold are eliminated. As another implementable method, the feature scores of all features can be sorted, and then the first e features are taken and placed in the fourth feature subset, and the remaining features are eliminated, where e can be an integer greater than or equal to 1.

[0108] The fourth feature subset can be accurately and simply obtained in the above manner, thus laying the foundation for obtaining the final feature.

[0109] Figure 6 The flowchart of using the embedded feature selection method to perform feature screening on the first-stage feature subset according to an embodiment of the present disclosure is schematically shown.

[0110] Operation S232 uses the embedded feature selection method to perform feature screening on the first-stage feature subset, including operation S2321.

[0111] In operation S2321, a gradient boosting algorithm (XGBoost algorithm) is used to iterate the first-stage feature subset, and a fifth feature subset is obtained during the iteration process.

[0112] It should be noted that the gradient boosting algorithm includes formula (11).

[0113] Fscore=|X| (11)

[0114] Among them, Fscore represents the importance score of the desired feature; X represents the set of leaf nodes that the desired feature is classified into. It can be understood that the gradient boosting algorithm is based on the tree model. In the process of building the tree model, each layer will select a feature segmentation point as a leaf node to maximize the gain value of the entire tree after segmentation. This means that the more times this feature is segmented in the tree, the greater the benefit it brings to the entire tree, and the more important the feature is. The following example shows that X represents the set of leaf nodes that the desired feature is classified into. For example, the set of leaf nodes can be expressed as {node 1, node 3, node 4}, which means that the desired feature is used as node 1, node 3 and node 4 respectively in the construction of the tree model. |X| represents the number of times the desired feature is used as a leaf node, which is 3 here, that is, the number of times the desired feature is selected for splitting.

[0115] Therefore, the importance score of each feature can be obtained by formula (11). As an implementable method, the importance score of each feature can be compared with the set fifth threshold, and the features with importance scores greater than the fifth threshold are placed in the fifth feature subset, and the features with importance scores less than the fifth threshold are eliminated. As another implementable method, the importance scores of all features can be sorted, and then the first g features are taken and placed in the fifth feature subset, and the remaining features are eliminated, where g can be an integer greater than or equal to 1.

[0116] The fifth feature subset can be accurately and simply obtained in the above manner, which also lays the foundation for obtaining the final feature.

[0117] Figure 7 The flowchart of determining data samples according to an embodiment of the present disclosure is schematically shown.

[0118] Operation S210 of determining the data sample includes operation S211 .

[0119] In operation S211, first data is obtained from the data storage system, and the first data is used as a data sample. It should be noted that the data storage system may be an account storage system of a bank, and the account storage system of the bank stores the account opening information of the customer, and the account opening information of the customer may be the first data. Thus, it is easy to obtain data samples, so that feature screening can be smoothly performed.

[0120] Figure 8 The flowchart of determining data samples according to an embodiment of the present disclosure is schematically shown.

[0121] Operation S210 of determining the data sample includes operations S212 to S213.

[0122] In operation S212, first data is acquired from a data storage system.

[0123] In operation S213, the abnormal feature value in the first data is regressed to obtain the second data, and the second data is used as a data sample. It can be understood that the data storage system can be an account storage system of a bank, and the account opening information of the customer is stored in the account storage system of the bank, and the account opening information of the customer can be the first data. Therefore, the first data includes but is not limited to features such as name, age, work unit, home address, telephone number, monthly income, and balance, and these features may have abnormal features. The values ​​under the abnormal features are abnormal feature values. For example, the missing or incorrectly filled values ​​under the feature work unit can be understood as the feature work unit being an abnormal feature. Among them, regressing the abnormal feature value can be understood as filling in the abnormal feature value or correcting the abnormal feature value.

[0124] As an implementable manner, the abnormal feature value may include a missing value, and regressing the abnormal feature value in the first data includes filling the missing value with the mean or median of the data under the abnormal feature. Thus, a more complete second data can be obtained, that is, a more complete data sample can be obtained, so that the final feature selected can more accurately determine whether the bank customer is about to be lost.

[0125] As another feasible method, the abnormal feature value may include a missing value, and regressing the abnormal feature value in the first data includes determining the missing value by induction. Thus, a more complete second data can also be obtained, that is, a more complete data sample can be obtained, so that the final feature selected can more accurately determine whether the bank customer is about to be lost.

[0126] Fig. 9 The flowchart of determining data samples according to an embodiment of the present disclosure is schematically shown.

[0127] Operation S210 of determining the data sample includes operations S214 to S216.

[0128] In operation S214, first data is obtained from multiple data storage systems respectively. It is understandable that a bank may have multiple data storage systems to store customer account opening information, and the customer account opening information may be used as the first data, and the first data may come from multiple data storage systems.

[0129] In operation S215, the outliers in the first data are regressed to obtain the second data. The outliers in the first data from the plurality of data storage systems may be regressed respectively, in other words, the outlier feature values ​​in the first data may be filled or corrected respectively, thereby obtaining the second data corresponding to the first data.

[0130] In operation S216, the second data obtained from different data storage systems are integrated to form a data sample. The second data from different data storage systems can be integrated to form a data sample. The second data integration can be understood as summarizing the second data into a matrix table, but the integration method is not limited to this. In this way, it is easy to obtain data samples, so that feature screening can be smoothly performed.

[0131] like Fig. 9 As shown, operation S210 of determining the data sample further includes operation S217.

[0132] In operation S217, a standardization method is used to perform standardization conversion on the integrated second data, and the data after standardization conversion is a data sample. The standardization conversion of the second data can process the data in the second data into data under the same dimension, which is convenient for calculation and modeling.

[0133] Furthermore, the standardization method may include standard deviation standardization, deviation standardization or Log function conversion, but is not limited thereto. Thus, the integrated second data may be standardized and converted by standard deviation standardization, deviation standardization or Log function conversion.

[0134] The following will be based on Figure 1 The scene described by Figure 10-11 The early warning method for bank customer churn in the disclosed embodiment is described in detail.

[0135] Fig.10 The flowchart of the early warning method for bank customer churn according to the embodiment of the present disclosure is schematically shown.

[0136] The early warning method for bank customer loss includes operations S310 to S340.

[0137] In operation S310 , the feature screening method for data as described above is performed to obtain final features.

[0138] In operation S320, the final features are input into the model as an input layer. The model can be understood as a model that applies the early warning method of bank customer loss, and the final features are features that can be used to determine whether bank customers are about to lose, such as monthly salary, monthly expenditure and other features.

[0139] In operation S330, the feature data of the input layer is converted into intermediate data. For example, the feature data of the input layer may be a value of monthly salary and / or a value of monthly expenditure, etc. The above values ​​may be converted into intermediate data to facilitate computer recognition and calculation. For example, the intermediate data may be data in a unified dimension.

[0140] In operation S340 , the intermediate data is converted into a final result, wherein the final result is used to characterize the customer as a customer prone to churn or a customer not prone to churn.

[0141] According to the bank customer loss early warning method of the embodiment of the present disclosure, it is possible to introduce the feature screening method for data into the bank customer loss early warning method, and then it is possible to predict which of the bank's customers are stable customers and which are customers that are about to be lost, so that the bank can retain the customers that are about to be lost in a targeted manner and improve the bank's service level and performance.

[0142] Fig.11 The flowchart of converting intermediate data into final results according to an embodiment of the present disclosure is schematically shown.

[0143] S340 converts the intermediate data into the final result, including operations S341 to S342.

[0144] In operation S341, a sigmoid function is used, and the sigmoid function calculates a mapping value between (0, 1).

[0145] In operation S342, the mapped value is compared with a standard threshold to obtain a final result.

[0146] Therefore, the mapping value can be obtained by applying the sigmoid function, and the final result can be obtained by comparing the mapping value with the standard threshold, so that it is easy to know whether the customer is a customer who is prone to churn or a customer who is not prone to churn.

[0147] Based on the above-mentioned feature screening method for data, the present disclosure also provides a device. Fig.12 The device 10 is described in detail.

[0148] Fig.12 The structural block diagram of the device 10 according to an embodiment of the present disclosure is schematically shown.

[0149] The device 10 comprises a data determination module 1 , a first screening module 2 and a second screening module 3 .

[0150] The data determination module 1 is used to perform operation S210: determine a data sample, determine a data sample that needs to be screened, and the data sample includes multiple features.

[0151] The first screening module 2 is used to perform operation S220: screening data in the first stage, the first stage screening data includes: using the maximum relevance-minimum redundancy algorithm (MRMR algorithm) to perform feature screening on the determined data samples to obtain a first feature subset; using the feature weight algorithm (Relief algorithm) to perform feature screening on the determined data samples to obtain a second feature subset; using the T-test algorithm (T-test algorithm) to perform feature screening on the determined data samples to obtain a third feature subset; and obtaining a union of the first feature subset, the second feature subset and the third feature subset;

[0152] The second screening module 3 is used to perform operation S230: second-stage screening data, the second-stage screening data including: using a packaged feature selection method to perform feature screening on the first-stage feature subset to obtain a fourth feature subset; using an embedded feature selection method to perform feature screening on the first-stage feature subset to obtain a fifth feature subset; and obtaining a final feature by taking the union of the fourth feature subset and the fifth feature subset.

[0153] Since the above-mentioned device is arranged based on the feature screening method for data, the beneficial effects of the above-mentioned device are the same as those of the feature screening method for data, and will not be described in detail here.

[0154] In addition, according to an embodiment of the present disclosure, any multiple modules among the data determination module 1, the first screening module 2 and the second screening module 3 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module.

[0155] According to an embodiment of the present disclosure, at least one of the data determination module 1, the first screening module 2, and the second screening module 3 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them.

[0156] Alternatively, at least one of the data determination module 1 , the first screening module 2 and the second screening module 3 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be executed.

[0157] Fig.13A block diagram of an electronic device suitable for implementing a feature screening method for data and an early warning method for bank customer churn according to an embodiment of the present disclosure is schematically shown.

[0158] like Fig.13 As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0159] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0160] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the I / O interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage portion 908 as needed.

[0161] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0162] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus, or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0163] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method of the embodiment of the present disclosure.

[0164] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 901. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0165] In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0166] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0167] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0168] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0169] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0170] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for feature screening of data, characterized in that: include: Determining a data sample, wherein the data sample includes a plurality of features; The first stage of data screening includes: using a maximum relevance-minimum redundancy algorithm, a feature weight algorithm, and a T-test algorithm to perform feature screening on the data samples respectively, and obtaining a first stage feature subset by combining the screening results; The second stage screening data includes: Using the package feature selection method to perform feature screening on the first-stage feature subset to obtain a fourth feature subset; Using an embedded feature selection method to perform feature screening on the first-stage feature subset to obtain a fifth feature subset; and Obtaining the final features according to the fourth feature subset and the fifth feature subset; The method of obtaining the final feature according to the fourth feature subset and the fifth feature subset includes calculating the importance score of each feature according to the indicator function, arranging the importance scores in ascending or descending order, and obtaining N final The final features, N final is an integer greater than or equal to 1, and the indicator function includes the following formula: Among them, f i represents each feature; when m=1, it represents the model of the packaged feature selection method, and when m=2, it represents the model of the embedded feature selection method; or, when m=2, it represents the model of the packaged feature selection method, and when m=1, it represents the model of the embedded feature selection method; when m represents the model of the packaged feature selection method, t i represents the number of the fourth feature subset. When m represents the model of the embedded feature selection method, t i Represents the number of the fifth feature subset; acc m,k represents the accuracy of model m at the kth run; N final Represents the number of final features; r(f i ,m,k) represents the feature f of model m at the kth run i The maximum rank of the features in each model m is recorded as .

2. The feature screening method for data according to claim 1, characterized in that: The method of using the packaged feature selection method to perform feature screening on the first-stage feature subset includes iterating the first-stage feature subset using a feature sorting algorithm, and obtaining a fourth feature subset during the iteration process.

3. The feature screening method for data according to claim 1, characterized in that: The use of the embedded feature selection method to perform feature screening on the first-stage feature subset includes iterating the first-stage feature subset using a gradient enhancement algorithm, and obtaining a fifth feature subset during the iteration process.

4. The feature screening method for data according to claim 1, characterized in that: The determining data sample comprises: Acquire first data from a data storage system, and use the first data as the data sample.

5. The feature screening method for data according to claim 1, characterized in that: The determining data sample comprises: Acquire first data from a data storage system; and The abnormal feature values ​​in the first data are regressed to obtain second data, and the second data are used as the data sample.

6. The feature screening method for data according to claim 5, characterized in that: The abnormal feature value includes a missing value, and regressing the abnormal feature value in the first data includes filling the missing value with the mean or median of the data under the abnormal feature or determining the missing value by induction.

7. The feature screening method for data according to claim 1, characterized in that: The determining data sample comprises: Acquire first data from a plurality of data storage systems respectively; Regressing the outliers in the first data to obtain second data; and The second data obtained from different data storage systems are integrated to form the data sample.

8. The method for feature screening of data according to claim 7, characterized in that: The determining of the data sample further comprises: The integrated second data is subjected to standardization conversion by using a standardization method, and the data subjected to the standardization conversion is the data sample.

9. A bank customer loss early warning method, characterized in that: include: Execute the feature screening method for data according to any one of claims 1 to 8 to obtain final features; Inputting the final features into the model as an input layer; Converting the feature data input into the input layer into intermediate data; as well as The intermediate data is converted into a final result, wherein the final result is used to characterize the customer as a churn-prone customer or a non-churn-prone customer.

10. The early warning method for bank customer loss according to claim 9, characterized in that: The converting of the intermediate data into the final result comprises: Using a sigmoid function, the sigmoid function calculates a mapping value between (0, 1); and The mapping value is compared with a standard threshold to obtain the final result.

11. A feature screening device for data, characterized in that: include: A data determination module is used to determine a data sample, and determine a data sample that needs to be screened, wherein the data sample includes multiple features; The first screening module is used to perform the first stage of data screening, wherein the first stage of data screening includes: using the maximum relevance-minimum redundancy algorithm, the feature weight algorithm and the T-test algorithm to perform feature screening on the data samples respectively, and obtaining the first stage feature subset by merging the screening results; The second screening module is used to perform the second stage of screening data, and the second stage of screening data includes: Using the package feature selection method to perform feature screening on the first-stage feature subset to obtain a fourth feature subset; Using an embedded feature selection method to perform feature screening on the first-stage feature subset to obtain a fifth feature subset; and Obtaining the final features according to the fourth feature subset and the fifth feature subset; The method of obtaining the final feature according to the fourth feature subset and the fifth feature subset includes calculating the importance score of each feature according to the indicator function, arranging the importance scores in ascending or descending order, and obtaining N final The final features, N final is an integer greater than or equal to 1, and the indicator function includes the following formula: Among them, f i represents each feature; when m=1, it represents the model of the packaged feature selection method, and when m=2, it represents the model of the embedded feature selection method; or, when m=2, it represents the model of the packaged feature selection method, and when m=1, it represents the model of the embedded feature selection method; when m represents the model of the packaged feature selection method, t i represents the number of the fourth feature subset. When m represents the model of the embedded feature selection method, t i Represents the number of the fifth feature subset; acc m,k represents the accuracy of model m at the kth run; N final Represents the number of final features; r(f i ,m,k) represents the feature f of model m at the kth run i The maximum rank of the features in each model m is recorded as .

12. An electronic device, characterized in that: include: one or more processors; One or more memories, used to store executable instructions, which, when executed by the processor, implement the method according to any one of claims 1 to 8.

13. A computer-readable storage medium, characterized in that: The storage medium stores executable instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 8.

14. A computer program, characterized in that The computer program includes one or more executable instructions, and when the executable instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images