Credit risk control model feature derivation method and system conforming to business logic
By using the feature derivation method in the credit risk control model, we build and screen out a new set of derived features that conforms to business logic, solving the problem of insufficient feature interpretability in the existing technology and improving the model's predictive ability and decision-making transparency.
Patent Information
- Application Number
- CN202411292005.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2025-01-24
AI Technical Summary
The existing credit risk control evaluation model fails to pay effective attention to the interpretability of features during the feature derivation process, resulting in a lack of transparency and credibility in decision-making analysis.
Propose a credit risk control model feature derivation method that conforms to business logic. By obtaining a sample set of marked users, select the business dimension fields in the credit report to preset conditions, build a derived feature set, and derive and quality through the full sample feature Evaluation filters out new derived feature sets that conform to business logic.
Realize the characteristics of automated batch derivatives, improve the predictive ability of the credit risk control model, and provide strong technical support to ensure the explanatoryness of the characteristics and the consistency of business logic.
Smart Images
Figure CN119359426A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of credit risk control technology, and specifically to a method and system for deriving credit risk control model characteristics that conform to business logic. Background Art
[0002] Credit risk control refers to the empowerment and expansion of traditional credit business through a series of technical means and measures in the context of Internet finance. With the rapid development of big data and artificial intelligence technology, financial institutions can obtain and process more and more comprehensive customer information, and improve the accuracy and efficiency of risk identification by building risk assessment models and applying machine learning algorithms. Among them, feature engineering is an important concept in machine learning technology and modeling. It refers to extracting or creating new features from complex and messy raw data to improve the predictive ability of the model.
[0003] In the feature derivation stage, we should not only pay attention to the IV (Information Value) effect value of the derived features, but also pay attention to the interpretability of the features. Not only model personnel, but also strategy personnel often use high-value derived features to support risk control decision analysis. Summary of the invention
[0004] The present application provides a credit risk control model feature derivation method and system that conforms to business logic, which can solve the technical problem in the prior art that the derived features in the credit risk control assessment feature engineering do not pay attention to the feature interpretability.
[0005] In a first aspect, the present application provides a credit risk control model feature derivation method that conforms to business logic, comprising the following steps:
[0006] Get a sample set of all marked users within a preset time range;
[0007] Select the business dimension fields in the credit report to preset conditions, obtain the preset function conditions, and the repayment information fields to preset targets, obtain the preset function targets, and select the statistical function;
[0008] Constructing a derived feature set for all samples in the sample set based on preset function conditions, function targets and selected statistical functions;
[0009] Traverse the derived feature set, perform feature derivation of all samples on each derived feature, and generate a new derived feature set;
[0010] Performing quality assessment on the new derived feature set to obtain quality assessment indicators;
[0011] According to the quality evaluation index, a new derived feature set that meets the business logic is screened from the new derived feature set.
[0012] In combination with the first aspect, in one implementation, the business dimension field in the selected credit report is used for condition preset to obtain the preset function condition, the repayment information field is used for target preset to obtain the preset function target, and the statistical function is selected:
[0013] The business dimensions include basic personal information, credit inquiry information, loan disbursement information, repayment information, and special transaction information;
[0014] The statistical function is a sum function, a counting function, an average function or a standard deviation function.
[0015] In combination with the first aspect, in one implementation, constructing a derived feature set for all samples in the sample set based on a preset function condition, a function target, and a selected statistical function specifically includes the following steps:
[0016] Perform binning preprocessing on the character class variables in the preset function conditions, perform binning preprocessing on the numerical class variables in the preset function conditions, and obtain the character class variables and numerical class variables after binning preprocessing;
[0017] A derived feature set is constructed based on the character variables and numerical variables preprocessed by binning, the preset function objectives and the selected statistical functions.
[0018] In combination with the first aspect, in one implementation, the character variables and numerical variables preprocessed based on binning, the preset function target and the selected statistical function are used to construct a derived feature set, and a single derived feature is shown as follows:
[0019] [condition1, condition2, target, function]
[0020] Among them, condition1 is the character variable after binning preprocessing, condition2 is the numeric variable after binning preprocessing, target is the preset function target, and function is the selected statistical function.
[0021] In combination with the first aspect, in one implementation, traversing the derived feature set, performing feature derivation of all samples on each derived feature, and generating a new derived feature set, performing feature derivation of all samples on each derived feature, specifically includes the following steps:
[0022] Perform grouping operations according to the conditions of condition1 and condition2, calculate the function with the function target as target, and superimpose the features of these four dimensions to form a new derived feature.
[0023] In combination with the first aspect, in one implementation, performing quality assessment on the new derived feature set to obtain a quality assessment index specifically includes the following steps:
[0024] Match the ID card number, mobile phone number, and application date of the marked user with the marking value and credit report corresponding to the user sample;
[0025] Combined with the tag value of each sample, the quality of the new derived feature set after information matching is evaluated to obtain the quality evaluation index.
[0026] In combination with the first aspect, in one implementation, the step of selecting a new derived feature set that conforms to the business logic from the new derived feature set according to the quality evaluation index specifically includes the following steps:
[0027] Count the evaluation indicators of the new derived feature set after information matching, perform a first round of screening on the new derived feature set after information matching according to the thresholds of the missing rate and the single value rate, and obtain the new derived feature set screened in the first round;
[0028] For the new derived feature set screened in the first round, perform equal frequency binning, calculate the feature stability index, and obtain the new derived feature set after the second round of screening;
[0029] For the new derived feature set screened out in the second round, business logic explainability analysis is performed to screen and obtain the new derived feature set that ultimately conforms to the business logic.
[0030] In combination with the first aspect, in one implementation, the business logic interpretability analysis is performed on the new derived feature set screened out in the second round, and the new derived feature set that finally conforms to the business logic is screened and obtained, specifically including the following steps:
[0031] For the new derived feature sets screened in the second round, select the required number from top to bottom, analyze whether they have business logic interpretability based on the monotonicity of binning, and screen to obtain new derived feature sets that meet logical interpretability;
[0032] The Pearson correlations of the features in the new derived feature set that meet the logical interpretability are calculated, and the features whose Pearson correlations do not meet the requirements are eliminated to obtain the final new derived feature set.
[0033] In the second aspect, the present application provides a credit risk control model feature derivation system that conforms to business logic, including:
[0034] A sample set acquisition module is used to acquire a sample set of all marked users within a preset time range;
[0035] The statistical function preset module is used to select the business dimension fields in the credit report for condition preset, obtain the preset function conditions, the repayment information fields for target preset, obtain the preset function targets, and select the statistical function;
[0036] A derived feature set construction module, which is in communication with the sample set acquisition module and the statistical function preset module, and is used to construct a derived feature set for all samples in the sample set based on preset function conditions, function targets and selected statistical functions;
[0037] A new derived feature set generation module is connected to the derived feature set construction module for traversing the derived feature set, performing full sample feature derivation on each derived feature, and generating a new derived feature set;
[0038] A feature set quality assessment module, which is in communication with the new derived feature set generation module and is used to perform quality assessment on the new derived feature set and obtain a quality assessment index;
[0039] The feature screening and acquisition module is in communication connection with the feature set quality assessment module and is used to screen the new derived feature set that conforms to the business logic from the new derived feature set according to the quality assessment index.
[0040] In conjunction with the second aspect, in one implementation, the derived feature set construction module includes:
[0041] A preprocessing unit, used to perform binning preprocessing on the character class variables in the preset function conditions, and to perform binning preprocessing on the numerical class variables in the preset function conditions, and to obtain the character class variables and numerical class variables after binning preprocessing;
[0042] The derived feature set construction unit is in communication connection with the preprocessing unit and is used to construct a derived feature set based on the character class variables and the numerical class variables after the binning preprocessing, the preset function target and the selected statistical function.
[0043] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:
[0044] This application proposes a credit risk control model feature derivation method that conforms to business logic. It adopts feature derivation to construct a derived feature set, and then performs quality assessment screening to obtain high-quality model features that can be automatically batch derived and conform to business logic for building a credit risk control assessment model. This is conducive to improving the model prediction ability of the credit risk control model and providing strong technical support for credit risk control. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1A flowchart of a method for deriving characteristics of a credit risk control model that complies with business logic for this application;
[0046] Figure 2 This is a schematic diagram of the functional modules of the credit risk control model feature derivation system that complies with the business logic of this application. DETAILED DESCRIPTION
[0047] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0048] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.
[0049] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.
[0050] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.
[0051] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.
[0052] First, some technical terms in the present application are explained to facilitate those skilled in the art to understand the present application.
[0053] IV: Information Value
[0054] PSl: Population Stability Index, population stability index;
[0055] The Combs method is a rule base reduction method for writing fuzzy logic rules described by William E. Combs in 1997. Its purpose is to prevent the combinatorial explosion of fuzzy logic rules.
[0056] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.
[0057] In a first aspect, the present application provides a credit risk control model feature derivation method that conforms to business logic, comprising the following steps:
[0058] Step S1: Obtain a sample set of all marked users within a preset time range;
[0059] Step S2: Select the business dimension field in the credit report to preset conditions, obtain the preset function conditions, preset the target for the repayment information field, obtain the preset function target, and select the statistical function;
[0060] Step S3: constructing a derived feature set for all samples in the sample set based on preset function conditions, function targets and selected statistical functions;
[0061] Step S4: traverse the derived feature set, perform feature derivation of all samples on each derived feature, and generate a new derived feature set;
[0062] Step S5: Perform quality assessment on the new derived feature set to obtain quality assessment indicators;
[0063] Step S6: According to the quality evaluation index, a new derived feature set that meets the business logic is screened from the new derived feature set.
[0064] This application proposes a credit risk control model feature derivation method that conforms to business logic. It adopts feature derivation to construct a derived feature set, and then performs quality assessment screening to obtain high-quality model features that can be automatically batch derived and conform to business logic for building a credit risk control assessment model. This is conducive to improving the model prediction ability of the credit risk control model and providing strong technical support for credit risk control.
[0065] In one embodiment, in the step S1: obtaining a sample set of all marked users within a preset time range, the data in the sample set includes the user's basic information, application time, mark value, credit channel information, credit report and other pre-loan data information.
[0066] In one embodiment, the step S2: select the business dimension field in the credit report to preset the condition, obtain the preset function condition, preset the repayment information field to preset the target, obtain the preset function target, and select the statistical function:
[0067] The business dimensions are commonly used business dimensions of credit reports, such as basic personal information, credit inquiry information, loan disbursement information, repayment information and special transaction information;
[0068] The selected statistical functions are a sum function, a count function, a mean value function and a standard deviation function.
[0069] In one embodiment, the step S3: constructing a derived feature set for all samples in the sample set based on a preset function condition, a function target and a selected statistical function, specifically comprises the following steps:
[0070] Step S31: pre-process the character class variables in the preset function conditions by binning, and pre-process the numerical class variables in the preset function conditions by binning, and obtain the character class variables and numerical class variables after binning pre-processing; specifically, pre-process the character class variables in the preset conditions (condition), such as city, marital status, etc., by binning, generally divided into 3-5 boxes; and pre-process the numerical class variables in the preset conditions (condition), such as age, the number of queries in the past 30 days, etc., by equal frequency binning according to frequency, generally divided into 5 boxes;
[0071] Step S32: construct a derived feature set based on the character variables and numerical variables preprocessed by binning, the preset function target and the selected statistical function, wherein a single derived feature is shown in the following formula:
[0072] [condition1, condition2, target, function]
[0073] Among them, condition1 is the character variable after binning preprocessing, condition2 is the numeric variable after binning preprocessing, target is the preset function target, and function is the selected statistical function.
[0074] The derived feature set is a set of derived features of all samples in the sample set.
[0075] In one embodiment, the present application adopts a cross-traversal search method to exhaustively enumerate tens of millions of logical combinations. The step S4: traverses the derived feature set, performs feature derivation of the entire sample for each derived feature, generates a new derived feature set, and performs feature derivation of the entire sample for each derived feature, specifically including the following steps:
[0076] Perform grouping operations according to the conditions of condition1 and condition2, calculate the function whose target is target, and superimpose the features of these four dimensions to form a new derived feature. The superposition in this step is the process of calculating the value of the target target using the specified function funtion under the classification conditions of condition1 and condition2.
[0077] Complete the calculation of the derived feature value of a comb, perform feature derivation of all samples for each comb, traverse all combs (derivative features) in combs (derivative feature set), perform the same comb logic calculation on the credit reports of all samples (i.e., feature superposition of four dimensions), and obtain a new derived feature set.
[0078] In one embodiment, the step S5: performing quality assessment on the new derived feature set to obtain a quality assessment index specifically includes the following steps:
[0079] Step S51: Match the ID card number, collection number, application date and credit tag value of the marked user with the credit report. This step can be implemented before step S1; wherein the credit tag value is manually defined according to the business situation, and the tag value includes 1 and 0. The business situation refers to the marking of good and bad customers in the credit business. Customers who are overdue within a certain period of time are recorded as 1 (i.e., good customers), and customers who are not overdue within a certain period of time are recorded as 0 (i.e., bad customers). The business situation refers to the credit business situation of the customer in terms of how many days they are overdue within a few months;
[0080] Step S52: Combine the tag value of each sample, and perform quality assessment on the new derived feature set after information matching to obtain quality assessment indicators, which include:
[0081] Missing rate: the ratio of samples with missing data (i.e., data cannot be defined or is empty) to the total number of samples, which measures the completeness and coverage of the data;
[0082] Extreme value: the maximum and minimum values among all sample values;
[0083] Maximum single value rate: the proportion of the single value that appears most times (i.e. the mode) to the total number of samples;
[0084] PSI value: Divide the sample into 5 boxes with equal frequency to calculate the characteristic stability index PSI value;
[0085] Monthly IV values and overall IV values: Calculate the characteristic IV values by month and overall.
[0086] Monotonicity of bin sorting: After the samples are divided into 5 bins with equal frequency, is the bin bad debt rate (the proportion of the marked value 1) monotonic?
[0087] Business explainability: Whether the new derived features are explainable, and whether the changes in the bad debt rate of each bin (the proportion of the mark value of 1) are consistent with the business logic.
[0088] In one embodiment, the method of selecting a new derived feature set that meets the business logic from the new derived feature set according to the quality evaluation index specifically includes the following steps:
[0089] Count the evaluation indicators of the new derived feature set after information matching, and perform a first round of screening on the new derived feature set after information matching according to the thresholds of missing rate and single value rate (usually 80%-90%) to obtain the new derived feature set screened in the first round;
[0090] For the newly derived feature set screened out in the first round, perform equal frequency binning of each feature into 5 points, then calculate the IV value and PSI value, remove the features with a PSI value greater than 0.2, sort them from high to low according to the IV value, perform the second round of screening, and obtain the newly derived feature set screened out in the second round;
[0091] For the new derived feature sets screened out in the second round, the required number is selected from top to bottom, and combined with the monotonicity of binning, it is analyzed whether it has business logic interpretability. The mutual Pearson correlation of the selected feature sets is calculated, and the features with low IV and correlation above 0.8 are eliminated to obtain the final new derived feature set.
[0092] This application proposes a method for deriving credit risk control model features that conforms to business logic. It uses a cross-traversal search method to exhaustively enumerate tens of millions of logical combinations, and then verifies the expert business logic after screening based on indicators such as IV values. It effectively solves the limitations of experts' limited business experience and avoids the defects of random derivation of machine learning without business significance, and solves the limitations of the prior art credit risk control model feature derivation in terms of interpretability.
[0093] Second, as Figure 2 As shown, the present application provides a credit risk control model feature derivation system that conforms to business logic, including a sample set acquisition module 100, a statistical function preset module 200, a derived feature set construction module 300, a new derived feature set generation module 400, a feature set quality assessment module 500 and a feature screening acquisition module 600. The sample set acquisition module 100 is used to acquire a sample set of all marked users within a preset time range; the statistical function preset module 200 is used to select the business dimension field in the credit report for condition preset, obtain the preset function condition, the repayment information field for target preset, obtain the preset function target, and select the statistical function; the derived feature set construction module 300 is communicated with the sample set acquisition module 100 and the statistical function preset module 200, and is used to construct a derived feature set for all samples in the sample set based on the preset function condition, function target and selected statistical function; the new derived feature set generation module 400 is communicated with the derived feature set construction module 300, and is used to traverse the derived feature set, perform full sample feature derivation for each derived feature, and generate a new derived feature set; the feature set quality assessment module 500 is communicated with the new derived feature set generation module 400, and is used to perform quality assessment on the new derived feature set and obtain quality assessment indicators; the feature screening acquisition module 600 is communicated with the feature set quality assessment module 500, and is used to screen a new derived feature set that meets the business logic from the new derived feature set according to the quality assessment indicators.
[0094] In one embodiment, the derived feature set construction module includes:
[0095] A preprocessing unit, used to perform binning preprocessing on the character class variables in the preset function conditions, and to perform binning preprocessing on the numerical class variables in the preset function conditions, and to obtain the character class variables and numerical class variables after binning preprocessing;
[0096] The derived feature set construction unit is in communication connection with the preprocessing unit and is used to construct a derived feature set based on the character class variables and the numerical class variables after the binning preprocessing, the preset function target and the selected statistical function.
[0097] Among them, the functional implementation of each module in the above-mentioned credit risk control model feature derivation system that conforms to business logic corresponds to the steps in the above-mentioned credit risk control model feature derivation method that conforms to business logic, and its functions and implementation process will not be repeated here one by one.
[0098] On the third aspect, an embodiment of the present application provides a credit risk control model feature derivative device that conforms to business logic. The credit risk control model feature derivative device that conforms to business logic can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.
[0099] The communication interface includes input / output (I / O) interface, physical interface and logical interface, etc., which are used to realize the interconnection of devices inside the credit risk control model feature derivative device that conforms to the business logic, and the interface used to realize the interconnection between the credit risk control model feature derivative device that conforms to the business logic and other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber optic interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.
[0100] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.
[0101] The processor may be a general-purpose processor, which may call the credit risk control model feature derivation program that conforms to the business logic stored in the memory, and execute the credit risk control model feature derivation method that conforms to the business logic provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the credit risk control model feature derivation program that conforms to the business logic is called may refer to the various embodiments of the credit risk control model feature derivation method that conforms to the business logic of the present application, which will not be repeated here.
[0102] In a fourth aspect, an embodiment of the present application also provides a readable storage medium.
[0103] The readable storage medium of the present application stores a credit risk control model feature derivation program that conforms to business logic, wherein when the credit risk control model feature derivation program that conforms to business logic is executed by a processor, the steps of the credit risk control model feature derivation method that conforms to business logic as described above are implemented.
[0104] Among them, the method implemented when the credit risk control model feature derivation program that conforms to business logic is executed can refer to the various embodiments of the credit risk control model feature derivation method that conforms to business logic of this application, and will not be repeated here.
[0105] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.
[0106] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.
[0107] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.
Claims
1. A credit risk control model feature derivation method that conforms to business logic, characterized in that: The following steps are involved: Get a sample set of all marked users within a preset time range; Select the business dimension fields in the credit report to preset conditions, obtain the preset function conditions, and the repayment information fields to preset targets, obtain the preset function targets, and select the statistical function; Constructing a derived feature set for all samples in the sample set based on preset function conditions, function targets and selected statistical functions; Traverse the derived feature set, perform feature derivation of all samples on each derived feature, and generate a new derived feature set; Performing quality assessment on the new derived feature set to obtain quality assessment indicators; According to the quality evaluation index, a new derived feature set that meets the business logic is screened from the new derived feature set.
2. The credit risk control model feature derivation method in accordance with business logic according to claim 1, characterized in that: The business dimension field in the selected credit report is used to preset conditions, obtain preset function conditions, the repayment information field is used to preset targets, obtain preset function targets, and the statistical function is selected: The business dimensions include basic personal information, credit inquiry information, loan disbursement information, repayment information, and special transaction information; The statistical function is a sum function, a counting function, an average function or a standard deviation function.
3. The credit risk control model feature derivation method in accordance with business logic according to claim 1, characterized in that: The step of constructing a derived feature set for all samples in the sample set based on a preset function condition, a function target, and a selected statistical function specifically includes the following steps: Perform binning preprocessing on the character class variables in the preset function conditions, perform binning preprocessing on the numerical class variables in the preset function conditions, and obtain the character class variables and numerical class variables after binning preprocessing; A derived feature set is constructed based on the character variables and numerical variables preprocessed by binning, the preset function objectives and the selected statistical functions.
4. The credit risk control model feature derivation method in accordance with business logic according to claim 1, characterized in that: In the derived feature set constructed based on the character variables and numerical variables preprocessed by binning, the preset function target and the selected statistical function, a single derived feature is shown as follows: [condition1, condition2, target, function] Among them, condition1 is the character variable after binning preprocessing, condition2 is the numeric variable after binning preprocessing, target is the preset function target, and function is the selected statistical function.
5. The credit risk control model feature derivation method in accordance with business logic according to claim 4, characterized in that: The traversing of the derived feature set, performing feature derivation of all samples on each derived feature, generating a new derived feature set, performing feature derivation of all samples on each derived feature, specifically includes the following steps: Perform grouping operations according to the conditions of condition1 and condition2, calculate the function with the function target as target, and superimpose the features of these four dimensions to form a new derived feature.
6. The credit risk control model feature derivation method in accordance with business logic according to claim 1, characterized in that: The quality assessment of the new derived feature set is performed to obtain a quality assessment index, specifically comprising the following steps: Match the ID card number, mobile phone number, and application date of the marked user with the marking value and credit report corresponding to the user sample; Combined with the tag value of each sample, the quality of the new derived feature set after information matching is evaluated to obtain the quality evaluation index.
7. The credit risk control model feature derivation method in accordance with business logic according to claim 6, characterized in that: The method of selecting a new derived feature set that meets the business logic from the new derived feature set according to the quality evaluation index specifically includes the following steps: Count the evaluation indicators of the new derived feature set after information matching, perform a first round of screening on the new derived feature set after information matching according to the thresholds of the missing rate and the single value rate, and obtain the new derived feature set screened in the first round; For the new derived feature set screened in the first round, perform equal frequency binning, calculate the feature stability index, and obtain the new derived feature set after the second round of screening; For the new derived feature set screened out in the second round, business logic explainability analysis is performed to screen and obtain the new derived feature set that ultimately conforms to the business logic.
8. The credit risk control model feature derivation method in accordance with business logic according to claim 7, characterized in that: The process of performing business logic interpretability analysis on the new derived feature set screened out in the second round, and screening and obtaining the new derived feature set that finally meets the business logic, specifically includes the following steps: For the new derived feature sets screened in the second round, select the required number from top to bottom, analyze whether they have business logic interpretability based on the monotonicity of binning, and screen to obtain new derived feature sets that meet logical interpretability; The Pearson correlations of the features in the new derived feature set that meet the logical interpretability are calculated, and the features whose Pearson correlations do not meet the requirements are eliminated to obtain the final new derived feature set.
9. A credit risk control model feature derivation system that complies with business logic, characterized in that: include: A sample set acquisition module is used to acquire a sample set of all marked users within a preset time range; The statistical function preset module is used to select the business dimension fields in the credit report for condition preset, obtain the preset function conditions, the repayment information fields for target preset, obtain the preset function targets, and select the statistical function; A derived feature set construction module, which is in communication with the sample set acquisition module and the statistical function preset module, and is used to construct a derived feature set for all samples in the sample set based on preset function conditions, function targets and selected statistical functions; A new derived feature set generation module is connected to the derived feature set construction module for traversing the derived feature set, performing full sample feature derivation on each derived feature, and generating a new derived feature set; A feature set quality assessment module, which is in communication with the new derived feature set generation module and is used to perform quality assessment on the new derived feature set and obtain a quality assessment index; The feature screening and acquisition module is in communication connection with the feature set quality assessment module, and is used to screen the new derived feature set that conforms to the business logic from the new derived feature set according to the quality assessment index.
10. The credit risk control model feature derivation system in accordance with business logic as claimed in claim 9, characterized in that: The derived feature set construction module includes: A preprocessing unit, used to perform binning preprocessing on the character class variables in the preset function conditions, and to perform binning preprocessing on the numerical class variables in the preset function conditions, and to obtain the character class variables and numerical class variables after binning preprocessing; The derived feature set construction unit is in communication connection with the preprocessing unit and is used to construct a derived feature set based on the character variables and numerical variables preprocessed by binning, the preset function target and the selected statistical function.