Scorecard modeling method and device based on five-fold cross validation

Abstract

The present invention discloses a scoring card modeling method and device based on five-fold cross validation, which defines a target variable; obtains original variable data to form a sample data set; performs binning on the original variable data, calculates the evidence weight value of each box variable, and the information value corresponding to the variable; uses the Pearson correlation coefficient to calculate the correlation between variables, and screens collinear variables; constructs a five-fold cross validation label based on the original variable data and the target variable; uses the five-fold cross validation label to randomly divide the sample data set into five parts, and uses the LightGBm algorithm to train the five-fold data respectively to obtain a scoring card model. The present invention can be well applied to the current credit scenario, fully mining the deep features in the massive third-party data to improve the distinguishing effect of the scoring card for different users.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a scoring card modeling method, apparatus, computer equipment, and storage medium based on five-fold cross-validation. Background Art

[0002] my country's consumer finance industry has a large room for growth.

[0003] Regarding the requirements of big data risk control, the data dimensions used by traditional financial risk control for risk control and the lack of risk control. At present, although financial big data uses more third-party data, which greatly enriches the available data dimensions of credit assessment and makes it possible to conduct credit assessment in an all-round and effective manner, it also puts forward higher requirements on the feature processing performance of the scoring card model. The feature set of financial credit institutions generally has the characteristics of large quantity and high latitude, but the scoring card should not focus on the data scale itself, but on the valuable features in these massive data. The traditional way of manual feature screening is no longer suitable for the current credit scenario and cannot fully explore the deep features in the massive third-party data. Although many companies have begun to try to use machine learning models (such as GBDT, AdaBoost, XGBoost, LightGBM, CatBoost, etc.) to replace the scoring card model, the division of the training set and the validation set still uses the scoring card to randomly divide them into two parts according to the labels, which easily causes the machine learning model to overfit. Generally, the scoring card model developed based on third-party data simply increases the dimension of the original input data without doing any other processing. In fact, the recall rates of different third-party data for modeling samples vary greatly. The use of traditional five-fold cross-validation may lead to large data gaps between different folds, ultimately resulting in poor output score ranking.

[0004] Regarding the problem in the above-mentioned related technologies that the traditional scoring card model cannot be well applied to the current credit scenario, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a scoring card modeling method, apparatus, computer device and storage medium based on five-fold cross-validation, so as to solve the problem that the traditional scoring card model in the related art is not well applicable to the current credit scenario.

[0006] In order to achieve the above object, a first aspect of an embodiment of the present invention provides a scoring card modeling method based on five-fold cross validation, comprising:

[0007] Define the target variable, define users whose overdue days exceed the preset number of days as negative samples, and define users whose overdue days do not exceed the preset number of days as positive samples;

[0008] Acquire original variable data, the original variable data including financial institution data and third-party data, to form a sample data set;

[0009] The original variable data is binned, and the weight of evidence of each box variable and the information value corresponding to the variable are calculated; the Pearson correlation coefficient is used to calculate the correlation between the variables, and the collinear variables are screened;

[0010] Constructing a five-fold cross validation label based on the original variable data and the target variable;

[0011] The sample data set is randomly divided into five parts using the five-fold cross-validation label, and the five-fold data are trained using the LightGBm algorithm to obtain a scorecard model.

[0012] Optionally, in a possible implementation manner of the first aspect, before the original variable data is binned, the method further includes:

[0013] The original variable data is subjected to data cleaning processing, wherein data cleaning includes missing value processing and outlier processing.

[0014] Optionally, in a possible implementation manner of the first aspect, constructing a five-fold cross validation label based on the original variable data and the target variable includes:

[0015] Marking missing values ​​for each set of feature values ​​in the third-party data and the financial institution data, respectively, wherein the third-party data or the financial institution data has at least one set of data sources, and each set of data sources has at least one set of feature values;

[0016] All missing value labels and the defined target variable are combined to form a multidimensional label as the five-fold cross-validation label.

[0017] Optionally, in a possible implementation manner of the first aspect, the method further includes:

[0018] Adjust the parameters of the model;

[0019] The model's discrimination ability, predictive ability and stability are evaluated according to the model's evaluation indicators, and an evaluation report is generated.

[0020] Optionally, in a possible implementation of the first aspect, the original variable data is binned, and the evidence weight value of each box variable and the information value corresponding to the variable are calculated; the correlation between the variables is calculated using the Pearson correlation coefficient, and the collinear variables are screened, including:

[0021] The weight of evidence value for each box variable is calculated as follows:

[0022]

[0023] Among them, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin;

[0024] The information value corresponding to the calculated variable is calculated as follows:

[0025]

[0026] Where N represents the number of variable bins, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin;

[0027] The Pearson correlation coefficient calculates the correlation between variables and is calculated as follows:

[0028]

[0029] Among them, r represents the Pearson correlation coefficient of variables X and Y, X k , k=1,2,…N represents the value of variable X in the kth sample, represents the mean of variable X, Y k , k=1,2,…N represents the value of variable Y in the kth sample, represents the mean of variable Y, and N is the total number of samples.

[0030] A second aspect of an embodiment of the present invention provides a scoring card modeling device based on five-fold cross validation, comprising:

[0031] A target variable definition module is used to define the target variable, define users whose overdue days exceed the preset number of days as negative samples, and define users whose overdue days do not exceed the preset number of days as positive samples;

[0032] The original variable acquisition module is used to acquire original variable data, wherein the original variable data includes financial institution data and third-party data, to form a sample data set;

[0033] The variable screening module is used to bin the original variable data, calculate the evidence weight value of each bin variable, and the information value corresponding to the variable; use the Pearson correlation coefficient to calculate the correlation between variables, and screen the collinear variables;

[0034] A validation label construction module, used to construct a five-fold cross-validation label based on the original variable data and the target variable;

[0035] The model training module is used to randomly divide the sample data set into five parts using the five-fold cross-validation labels, and train the five-fold data using the LightGBm algorithm to obtain a scorecard model.

[0036] Optionally, in a possible implementation manner of the second aspect, the device further includes:

[0037] The data cleaning module is used to perform data cleaning processing on the original variable data, wherein data cleaning includes missing value processing and outlier processing.

[0038] Optionally, in a possible implementation manner of the second aspect, the verification tag construction module further includes:

[0039] A missing value marking unit, used to mark missing values ​​for each set of characteristic values ​​in the third-party data and the financial institution data, wherein the third-party data or the financial institution data has at least one set of data sources, and each set of data sources has at least one set of characteristic values;

[0040] The validation label generation unit is used to combine all missing value labels and the defined target variable to form a multidimensional label as a five-fold cross-validation label.

[0041] According to a third aspect of an embodiment of the present invention, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0042] According to a fourth aspect of an embodiment of the present invention, a readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the steps of the method of the first aspect of the present invention and various possible designs of the first aspect.

[0043] The scoring card modeling method, device, computer equipment and storage medium based on five-fold cross validation provided by the present invention define the target variable, define the users whose overdue days exceed the preset days as negative samples, and define the users whose overdue days do not exceed the preset days as positive samples; obtain the original variable data, the original variable data includes financial institution data and third-party data, and form a sample data set; bin the original variable data, calculate the evidence weight value of each box variable, and the information value corresponding to the variable; use the Pearson correlation coefficient to calculate the correlation between variables, and screen the collinear variables; construct a five-fold cross validation label based on the original variable data and the target variable; use the five-fold cross validation label to randomly divide the sample data set into five parts, and use the LightGBm algorithm to train the five-fold data respectively to obtain a scoring card model. The present invention can be well applied to the current credit scenario, fully explore the deep features in the massive third-party data, so as to improve the distinguishing effect of the scoring card for different users. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of a scoring card modeling method based on five-fold cross validation provided in an embodiment of the present invention;

[0045] Figure 2 A structural diagram of a scoring card modeling device based on five-fold cross validation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0048] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0049] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0050] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0051] It should be understood that in the present invention, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A and B can be determined based on A. Determining B based on A does not mean determining B based only on A, but B can also be determined based on A and / or other information. A and B match when the similarity between A and B is greater than or equal to a preset threshold.

[0052] Depending on the context, "if" as used herein may be interpreted as "when" or "when" or "in response to determining" or "in response to detecting."

[0053] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0054] The present invention provides a scoring card modeling method based on five-fold cross validation, such as Figure 1 The flow chart shown includes:

[0055] Step S110: define target variables.

[0056] In this step, it is necessary to set the target variable in advance, that is, based on vintage analysis, observe the trend of average overdue payments in each month, determine the time span of the performance window, and define users whose overdue days during the performance period exceed the preset number of days (for example, 90 days) as negative samples (or defined as "bad users"), and define users whose overdue days during the performance period do not exceed the preset number of days (for example, 90 days) as positive samples (or defined as "good users").

[0057] Step S120: Acquire original variable data to form a sample data set.

[0058] In step S120, the original variable data mainly includes two aspects, one is financial institution data (including but not limited to the credit report of the People's Bank of China, bank flow, basic personal attribute information of the platform application, where the basic personal attribute information mainly includes name, ID number and mobile phone number); the other is third-party data (including but not limited to multiple lending behavior information, shopping information of major network platforms, third-party payment information and information about anti-fraud behavior). In the acquisition process, different acquisition methods are adopted for data transmission according to the confidentiality of the data. Some data (lower confidentiality or non-confidentiality) can be crawled online according to the web crawler technology, or some data (higher confidentiality) can be transmitted through the specific data interface of the designated financial institution or third party.

[0059] Step S130: bin the original variable data, calculate the evidence weight value of each box variable and the information value corresponding to the variable; use the Pearson correlation coefficient to calculate the correlation between the variables, and screen the collinear variables.

[0060] In this step, the following principles need to be followed when binning: 1. Categorical variables with fewer values ​​do not need to be binned; 2. The binning results need to be ordered; 3. The binning is balanced, and the data in the smallest bin is not less than 5%; 4. The monotonicity of the binning: Under strict requirements, the bad sample rate is monotonically related to the bin; when non-monotonic, it needs to be merged with the front bin and the back bin; 5. Not all bins are "good customers" or "bad customers". In the binning process, supervised binning or unsupervised binning can be adopted. Supervised binning (such as decision tree binning, optimal segment binning, etc.) can be combined with the target variable to maximize the information of the target variable in the feature; unsupervised binning (such as equidistant binning, cluster binning) has the advantage of simple calculation and the disadvantage of not being able to guarantee its rationality.

[0061] Specifically, in the risk control scorecard model, for business interpretability, the role of linear judgment variables is played, that is, when the feature variable x becomes larger and larger, the target variable y will also become larger and larger, but in reality there is often a nonlinear relationship between x and y. At this time, the WOE transformation can be used to calculate the evidence weight value of each box variable. The calculation formula of the evidence weight value is as follows:

[0062]

[0063] Among them, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin;

[0064] Calculate the weight of evidence WOE for the bin i Then, according to its weight of evidence value WOE i Calculate the information value IV corresponding to the variable, so as to evaluate the predictive ability of the variable. Generally speaking, the higher the information value IV, the stronger the predictive ability. IV<0.02 is an invalid feature; 0.02<=IV<0.1 is a weak feature; 0.1<=IV<=0.5 is an effective feature; IV>0.5 is a strong feature. The calculation formula of information value IV is as follows:

[0065]

[0066] Where N represents the number of variable bins, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin;

[0067] The Pearson correlation coefficient was used to calculate the correlation between variables, and the collinear variables were screened. For those with a Pearson correlation coefficient greater than 0.7, the variables with small information values ​​were deleted.

[0068] The calculation formula is as follows:

[0069]

[0070] Among them, r represents the Pearson correlation coefficient of variables X and Y, X k , k=1,2,…N represents the value of variable X in the kth sample, represents the mean of variable X, Y k , k=1,2,…N represents the value of variable Y in the kth sample, represents the mean of variable Y, and N is the total number of samples.

[0071] Step S140: construct a five-fold cross-validation label based on the original variable data and the target variable.

[0072] In step S140, the following two steps are mainly included:

[0073] Step 1: First, each set of characteristic values ​​in the third-party data and the financial institution data is marked with missing values, wherein the third-party data or the financial institution data has at least one set of data sources, and each set of data sources has at least one set of characteristic values.

[0074] Take an example to illustrate step 1: Assume that there are N third-party data involved in the modeling, represented as t1, t2, ..., tn, ..., tN, respectively. For the third-party data source tn, the data source provides Mn-dimensional features, represented as tn1, tn2, ..., tnmn, ..., tnMn, respectively. The Mn-dimensional features are marked with missing values, that is, the missing samples in the feature tnmn are marked as 1, otherwise they are marked as 0. The missing value label of the m-dimensional feature of the n-th third-party data source is represented as ynmn. Similarly, each feature under all financial institution data sources is marked with missing values.

[0075] Step 2:

[0076] Then all missing value labels are combined with the defined target variable to form a multidimensional label as the five-fold cross-validation label. The label of this sample is as follows:

[0077] (y11,y12,...,y1M1;

[0078] y21,y22,...,y2M2; ......,

[0079] y11,y12,...,yNM N ,

[0080] y).

[0081] Step S150: randomly divide the sample data set into five parts using the five-fold cross-validation labels, and train the five-fold data using the LightGBm algorithm to obtain a scorecard model.

[0082] In step S150, the sample is randomly divided into five parts based on the five-fold cross-validation label of step S140, and the proportion of all labels in each fold is ensured to be consistent with the overall sample. Then the five-fold data are trained using the LightGBm algorithm, and the mean of the five-fold prediction results is calculated as the final output, and the test set is used to verify the evaluation indicators of the trained model.

[0083] In one embodiment, before the original variable data is binned, the method further includes: performing data cleaning on the original variable data, wherein data cleaning includes missing value processing and outlier processing.

[0084] In this step, after obtaining the original variable data (financial institution data and third-party data), the fields of the financial institution itself and the third-party data fields are classified to understand the general situation of the data, including the missing values, outliers, average values, median values, maximum values, minimum values ​​and distribution of each field; and then the dirty data is cleaned. Among them, it is necessary to take data from the data pool according to the definition of the observation period and the performance period, and perform preliminary data cleaning and stability verification. Data cleaning includes user uniqueness check, missing value check, outlier check, etc. Stability verification mainly examines the stability of the variable in the time series, and the indicators measured include PSI, mean / variance, IV, etc.

[0085] In one embodiment, the method further includes: adjusting the parameters of the model, evaluating the distinguishing ability, predictive ability and stability of the model according to the evaluation indicators of the model, and generating an evaluation report, comparing it with other algorithm models, and drawing a conclusion on whether the model can be used.

[0086] In one embodiment, the method includes: selecting a system platform for deploying a scorecard model, deploying the scorecard model on the system platform, monitoring the information value IV, mean, PSI, AUC of the model, and regularly updating the model based on the monitoring.

[0087] The scoring card modeling method based on five-fold cross-validation provided by the present invention defines a target variable, defines users whose overdue days exceed a preset number of days as negative samples, and defines users whose overdue days do not exceed the preset number of days as positive samples; obtains original variable data, wherein the original variable data includes financial institution data and third-party data to form a sample data set; bins the original variable data, and calculates the evidence weight value of each box variable and the information value corresponding to the variable; uses the Pearson correlation coefficient to calculate the correlation between variables, and screens collinear variables; constructs a five-fold cross-validation label based on the original variable data and the target variable; uses the five-fold cross-validation label to randomly divide the sample data set into five parts, and trains the five-fold data using the LightGBm algorithm to obtain a scoring card model. The present invention can be well applied to the current credit scenario. By introducing a machine learning algorithm, the deep features in the massive third-party data can be fully mined to improve the distinguishing effect of the scoring card on different users. By adopting a five-fold cross-validation, the overfitting of the machine learning model can be avoided, making the final prediction result more stable and more robust. And by introducing the missing information of the data source into the five-fold label, the gap between the five-fold data is within a reasonable range, thereby better exerting the advantages of the machine learning algorithm in the scoring card modeling.

[0088] The embodiment of the present invention also provides a scoring card modeling device based on five-fold cross validation, such as Figure 2As shown, including:

[0089] A target variable definition module is used to define the target variable, define users whose overdue days exceed the preset number of days as negative samples, and define users whose overdue days do not exceed the preset number of days as positive samples;

[0090] The original variable acquisition module is used to acquire original variable data, wherein the original variable data includes financial institution data and third-party data, to form a sample data set;

[0091] The variable screening module is used to bin the original variable data, calculate the evidence weight value of each bin variable, and the information value corresponding to the variable; use the Pearson correlation coefficient to calculate the correlation between variables, and screen the collinear variables;

[0092] A validation label construction module, used to construct a five-fold cross-validation label based on the original variable data and the target variable;

[0093] The model training module is used to randomly divide the sample data set into five parts using the five-fold cross-validation labels, and train the five-fold data using the LightGBm algorithm to obtain a scorecard model.

[0094] In one embodiment, the apparatus further comprises:

[0095] The data cleaning module is used to perform data cleaning processing on the original variable data, wherein data cleaning includes missing value processing and outlier processing.

[0096] In one embodiment, the verification tag building module further includes:

[0097] A missing value marking unit, used to mark missing values ​​for each set of characteristic values ​​in the third-party data and the financial institution data, wherein the third-party data or the financial institution data has at least one set of data sources, and each set of data sources has at least one set of characteristic values;

[0098] The validation label generation unit is used to combine all missing value labels and the defined target variable to form a multidimensional label as a five-fold cross-validation label.

[0099] Among them, the readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general or special-purpose computer. For example, a readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the readable storage medium can also exist in a communication device as discrete components. The readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0100] The present invention also provides a program product, which includes an execution instruction, which is stored in a readable storage medium. At least one processor of a device can read the execution instruction from the readable storage medium, and at least one processor executes the execution instruction so that the device implements the methods provided in the above various embodiments.

[0101] In the above-mentioned terminal or server embodiment, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the present invention can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A scoring card modeling method based on five-fold cross validation, characterized in that: include: Define the target variable, define users whose overdue days exceed the preset number of days as negative samples, and define users whose overdue days do not exceed the preset number of days as positive samples; Acquire original variable data, the original variable data including financial institution data and third-party data, to form a sample data set; The original variable data is binned, and the weight of evidence of each box variable and the information value corresponding to the variable are calculated; the Pearson correlation coefficient is used to calculate the correlation between the variables, and the collinear variables are screened; Constructing a five-fold cross validation label based on the original variable data and the target variable; The sample data set is randomly divided into five parts using the five-fold cross-validation label, and the five-fold data are trained using the LightGBm algorithm to obtain a scorecard model; The original variable data is binned, and the weight of evidence for each variable and the information value corresponding to the variable are calculated; the Pearson correlation coefficient is used to calculate the correlation between variables, and the collinear variables are screened, including: The weight of evidence value for each box variable is calculated as follows: Among them, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin; The information value corresponding to the calculated variable is calculated as follows: Where N represents the number of variable bins, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin; The Pearson correlation coefficient calculates the correlation between variables and is calculated as follows: Among them, r represents the Pearson correlation coefficient of variables X and Y, X k , k=1,2,…N represents the value of variable X in the kth sample, represents the mean of variable X, Y k , k=1,2,…N represents the value of variable Y in the kth sample, represents the mean of variable Y, and N is the total number of samples.

2. The scoring card modeling method based on five-fold cross validation according to claim 1, characterized in that: Before the original variable data is binned, the method further includes: The original variable data is subjected to data cleaning processing, wherein data cleaning includes missing value processing and outlier processing.

3. The scoring card modeling method based on five-fold cross validation according to claim 1, characterized in that: Constructing a five-fold cross validation label based on the original variable data and the target variable, including: Marking missing values ​​for each set of feature values ​​in the third-party data and the financial institution data, respectively, wherein the third-party data or the financial institution data has at least one set of data sources, and each set of data sources has at least one set of feature values; All missing value labels and the defined target variable are combined to form a multidimensional label as the five-fold cross-validation label.

4. The scoring card modeling method based on five-fold cross validation according to claim 1, characterized in that: The method further comprises: Adjust the parameters of the model; The model's discrimination ability, predictive ability and stability are evaluated according to the model's evaluation indicators, and an evaluation report is generated.

5. A scoring card modeling device based on five-fold cross validation, characterized in that: include: A target variable definition module is used to define the target variable, define users whose overdue days exceed the preset number of days as negative samples, and define users whose overdue days do not exceed the preset number of days as positive samples; The original variable acquisition module is used to acquire original variable data, wherein the original variable data includes financial institution data and third-party data, to form a sample data set; The variable screening module is used to bin the original variable data, calculate the evidence weight value of each bin variable, and the information value corresponding to the variable; use the Pearson correlation coefficient to calculate the correlation between variables, and screen the collinear variables; A validation label construction module, used to construct a five-fold cross-validation label based on the original variable data and the target variable; A model training module is used to randomly divide the sample data set into five parts using the five-fold cross-validation labels, and train the five-fold data using the LightGBm algorithm to obtain a scorecard model; The original variable data is binned, and the evidence weight value of each box variable and the information value corresponding to the variable are calculated; The Pearson correlation coefficient was used to calculate the correlation between variables and screen the collinear variables, including: The weight of evidence value for each box variable is calculated as follows: Among them, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin; The information value corresponding to the calculated variable is calculated as follows: Where N represents the number of variable bins, WOE i represents the evidence weight value of the i-th bin, p i,good 、p i,bad Respectively represent the ratio of the number of positive and negative samples to the total number of samples in the i-th bin; The Pearson correlation coefficient calculates the correlation between variables and is calculated as follows: Among them, r represents the Pearson correlation coefficient of variables X and Y, X k , k=1,2,…N represents the value of variable X in the kth sample, represents the mean of variable X, Y k , k=1,2,…N represents the value of variable Y in the kth sample, represents the mean of variable Y, and N is the total number of samples.

6. The scoring card modeling device based on five-fold cross validation according to claim 5, characterized in that: The device further comprises: The data cleaning module is used to perform data cleaning processing on the original variable data, wherein data cleaning includes missing value processing and outlier processing.

7. The scoring card modeling device based on five-fold cross validation according to claim 5, characterized in that: The verification tag building module further includes: A missing value marking unit, used to mark missing values ​​for each set of characteristic values ​​in the third-party data and the financial institution data, wherein the third-party data or the financial institution data has at least one set of data sources, and each set of data sources has at least one set of characteristic values; The validation label generation unit is used to combine all missing value labels and the defined target variable to form a multidimensional label as a five-fold cross-validation label.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images