Small and micro subject credit evaluation method and system based on model fusion technology

Abstract

The invention relates to a micro subject credit evaluation method and system based on a model fusion technology, and the method comprises the steps: carrying out the preprocessing of obtained training data, and obtaining standard data; performing binning on each index feature in the standard data, screening out the index features with prediction capability based on an IV value, and screening out a second number of index features based on a stepwise regression principle and a correlation test; the evidence weight and the importance weight of the second number of index features are calculated, and an enterprise subject score card model and an enterprise owner score card model are constructed according to the evidence weight and the importance weight; and for different customer groups, on the basis of a weighted average algorithm, respectively fusing the enterprise subject score card model and the enterprise owner score card model to evaluate the credit level of the small and micro enterprises in the credit scene. According to the method and the device, the problem of poor model effect caused by insufficient evaluation data, evaluation standard applicability and modeling sample quantity in credit evaluation of small and micro enterprises in a credit and loan scene is solved.

Description

Technical Field

[0001] This application relates to the technical field of enterprise evaluation, and particularly to a credit evaluation method and system for small and micro entities based on model fusion technology. Background Art

[0002] The credit evaluation of small and micro enterprises is a comprehensive, objective, and fair evaluation of the credit status of small and micro enterprises and individual industrial and commercial households, aiming to reveal the credit risks of enterprises and provide important basis for the credit decision-making of financial institutions. The credit evaluation results of small and micro enterprises are mainly applied to multiple business scenarios such as the access approval of financial institutions, customer risk stratification, differential quota pricing, in-loan risk monitoring, renewal loan amount increase, and marketing poor customer screening after the enterprises apply for business with financial institutions.

[0003] In recent years, the state has emphasized increasing the investment in credit loans by strengthening the exploration and utilization of the credit information of small and micro enterprises, strengthening the capacity building of financial institutions such as marketing customer acquisition, credit approval, and risk management, and promoting the development of inclusive finance. Financial institutions such as banks and third-party service institutions such as credit reporting companies have also carried out research and application on the credit evaluation standards and models of small and micro enterprises, but there are still many problems at present:

[0004] Insufficient evaluation data. Compared with large enterprises, small and micro enterprises often lack complete financial information, and there may not be enough traces and records of their daily operations, such as using cash or the personal accounts of enterprise owners for daily settlement, and not inviting accounting firms to issue audit reports. In addition, small and micro enterprises generally have a short establishment period and lack loan history, resulting in less credit data and historical data of small and micro enterprises, making it difficult for financial institutions such as banks to collect and identify the effective information of small and micro enterprises.

[0005] Insufficient applicability of evaluation standards. The methods for financial institutions to evaluate the credit of large and medium-sized enterprises are relatively mature, mainly based on indicators such as the asset scale, profitability, and financial status of enterprises. However, small and micro enterprises often lack information such as complete financial reports, making it difficult for financial institutions to determine the credit risk level of enterprises through traditional evaluation means.

[0006] Insufficient modeling sample size. Generally, financial institutions have a small number of small and micro enterprise customers, and the real business samples available for quantitative model analysis are insufficient, making it difficult to achieve good model effects.

[0007] Currently, for the problems of insufficient evaluation data, insufficient applicability of evaluation standards, and poor model effects caused by insufficient modeling sample quantity in the credit evaluation of small and micro enterprises in related technologies, no effective solutions have been proposed. Summary of the Invention

[0008] The embodiments of the present application provide a credit evaluation method and system for small and micro entities based on model fusion technology, so as to at least solve the problems of insufficient evaluation data, applicability of evaluation criteria, and poor model effect caused by insufficient number of modeling samples in the credit evaluation of small and micro enterprises in the related art.

[0009] In a first aspect, the embodiments of the present application provide a credit evaluation method for small and micro entities based on model fusion technology, and the method includes:

[0010] Obtain training data, and preprocess the training data to obtain standard data;

[0011] Bin each index feature in the standard data to obtain a number of index bins for the index feature;

[0012] According to the index bins of the index feature, calculate the IV value of the index feature, and screen the index feature based on the IV value to obtain an index feature with predictive ability;

[0013] Based on the stepwise regression principle and correlation test, screen out a second number of index features from the index features with predictive ability;

[0014] Calculate the weight of evidence for each index bin in the second number of index features;

[0015] Based on the Logistic regression model, calculate the importance weight value of the second number of index features;

[0016] Based on the weight of evidence and the importance weight value, construct an enterprise entity and an enterprise owner scoring card model respectively;

[0017] For the small and micro enterprise customer group and the individual industrial and commercial household customer group, calculate the model results after the fusion of the enterprise entity scoring card model and the enterprise owner scoring card model respectively through the weighted average algorithm, and use them to evaluate the credit level of small and micro enterprises in the credit scenario.

[0018] In some embodiments, binning each index feature in the standard data to obtain a number of index bins for the index feature includes:

[0019] Sort the variable values of each index feature in the standard data, initialize each variable value as an index bin, traverse and calculate the chi-square value of adjacent two index bins, merge the two index bins with the smallest chi-square value, and repeat the traversal and merging until a preset number of index bins are obtained.

[0020] In some embodiments, according to the index bins of the index feature, calculate the IV value of the index feature, and screen the index feature based on the IV value to obtain an index feature with predictive ability includes:

[0021] Calculate the IV value of the index feature through the IV value calculation formula, where B is the number of enterprises that default in the i-th index bin, and Bis the number of all defaulting enterprises, G T is the number of enterprises that perform their obligations in the i-th index bin, and G i is the number of all enterprises that perform their obligations; T Based on the IV value, screen the index features, and select the index features with predictive ability whose IV value is greater than or equal to 0.02.

[0022] In some embodiments, based on the stepwise regression principle and correlation test, screening out the second quantity of index features from the index features with predictive ability includes:

[0023] Based on the stepwise regression principle, screen out the first quantity of index features from the index features with predictive ability;

[0024] Based on the correlation test, screen out the second quantity of index features from the first quantity of index features.

[0025] In some embodiments, based on the stepwise regression principle, screening out the first quantity of index features from the index features with predictive ability includes:

[0026] Using the Akaike information criterion AIC = 2k - 2ln(L) as the index of the stepwise regression principle, screen out the first quantity of index features from the index features with predictive ability, where k is the number of index features and L is the likelihood function.

[0027] In some embodiments, based on the correlation test, screening out the second quantity of index features from the first quantity of index features includes:

[0028] Calculate the correlation coefficients between the first quantity of index features through the correlation coefficient calculation formula

[0029] where X and Y are two different index features. If the correlation coefficient is greater than 0.5, then retain the index feature with the higher IV value among the two index features, and finally obtain the second quantity of index features.

[0030] In some embodiments, calculating the weight of evidence for each index bin in the second quantity of index features includes:

[0031] Calculate the weight of evidence for each index bin in the second quantity of index features through the coding formula where p i1 ​is the proportion of defaulting enterprises in the i-th index bin among all defaulting enterprises, p i0 is the proportion of compliant enterprises in the i-th index bin among all compliant enterprises, B i is the number of defaulting enterprises in the i-th index bin, B T is the number of all defaulting enterprises, G i is the number of compliant enterprises in the i-th index bin, G T is the number of all compliant enterprises.

[0032] In some embodiments, based on the Logistic regression model, calculating the importance weights of the second quantity of index features includes:

[0033] Based on the mathematical representation of the Logistic regression model Calculating the importance weights of the second quantity of index features, where p is the probability of enterprise default, x is the index feature value, θ is the importance weight of the index feature, and T represents the transpose of the matrix.

[0034] In some embodiments, based on the weight of evidence and the importance weights, enterprise entity and enterprise owner scorecard models are respectively constructed;

[0035] In some embodiments, for the small and micro enterprise customer group and the self-employed industrial and commercial household customer group, their enterprise entity scorecard models and enterprise owner scorecard models are respectively integrated to evaluate the credit level of small and micro enterprises in the credit scenario, including:

[0036] For the small and micro enterprise customer group and the self-employed industrial and commercial household customer group, the combined model results and model KS values of the enterprise entity scorecard model and the enterprise owner scorecard model are respectively calculated through the weighted average algorithm, and the weight combination of the enterprise entity scorecard model and the enterprise owner scorecard model when KS is the largest is selected as the optimal weight combination for the final model integration, which is used to evaluate the credit level of small and micro enterprises in the credit scenario;

[0037] Collect data according to the unified social credit code, owner name, owner ID number, and owner mobile phone number of the preset enterprise from several data sources. Among them, the data sources include the National Enterprise Credit Information Publicity System and the China Judgments Online, etc., and the data types include industrial and commercial data, judicial data, intellectual property data, multi-headed data, device behavior data, consumption data, etc.;

[0038] Clean the data and convert it into the data type that meets the requirements of the scorecard model;

[0039] Input the converted data into the scorecard model for scoring;

[0040] The scores of the scoring card models are fused according to the optimal weight combination to obtain the final score, which reflects the credit level of the preset enterprise in the credit scenario.

[0041] In a second aspect, the embodiments of the present application provide a credit evaluation system for small and micro entities based on model fusion technology. The system includes a data acquisition module, a feature screening module, a model construction module, and a model fusion module;

[0042] The data acquisition module is used to acquire training data and preprocess the training data to obtain standard data;

[0043] The feature screening module is used to bin each index feature in the standard data to obtain several index bins of the index feature; calculate the IV value of the index feature according to the index bin of the index feature, and screen the index feature based on the IV value to obtain index features with predictive ability; based on the stepwise regression principle and correlation test, screen out a second number of index features from the index features with predictive ability;

[0044] The model construction module is used to calculate the weight of evidence of each index bin in the second number of index features; calculate the importance weight value of the second number of index features based on the Logistic regression model; respectively construct an enterprise entity scoring card model and an enterprise owner scoring card model based on the weight of evidence and the importance weight value.

[0045] The model fusion module is used to calculate the model results and KS values after fusing different customer group enterprise entity scoring card models and enterprise owner scoring card models, select the weight combination when the KS value of the fused model is the largest, fuse the enterprise entity and enterprise owner scoring models of the small and micro enterprise customer group and the individual industrial and commercial household customer group, and evaluate the credit level of enterprise customers in the credit scenario.

[0046] Compared with the related art, an SME credit evaluation method and system based on model fusion technology provided by an embodiment of the present application preprocesses the acquired training data to obtain standard data; bins each index feature in the standard data to obtain several index bins of the index feature, screens out the index features with predictive ability according to the index bins, and then based on the stepwise regression principle and correlation test, screens out a second number of index features from them; calculates the weight of evidence of each index bin in the second number of index features, calculates the importance weight of the second number of index features based on the Logistic regression model, and constructs an enterprise entity scoring card model and an enterprise owner scoring card model respectively based on the weight of evidence and the importance weight; based on the entity nature, calculates the model results and KS values after the fusion of the enterprise entity and enterprise owner scoring card models for the SME customer group and the self-employed industrial and commercial household customer group respectively, and selects the optimal weight combination when the KS value of the fused model is the largest to obtain the final fused model, which is used to evaluate the credit level in the credit scenario of SME customers. It solves the problems of insufficient evaluation data, applicability of evaluation criteria, and poor model effect caused by insufficient number of modeling samples in the credit evaluation of SME customers in the credit scenario, realizes the acquisition of data from multiple sources, quantifies the credit evaluation criteria for SMEs, and constructs and optimizes the model through a large number of real samples to improve the accuracy of model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0048] Figure 1 is a flowchart of the steps of an SME credit evaluation method based on model fusion technology according to an embodiment of the present application;

[0049] Figure 2 is a structural block diagram of an SME credit evaluation system based on model fusion technology according to an embodiment of the present application;

[0050] Figure 3 is an internal structural schematic diagram of an electronic device according to an embodiment of the present application.

[0051] BRIEF DESCRIPTION OF THE DRAWINGS: 21, data acquisition module; 22, feature screening module; 23, model construction module; 24, model fusion module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be described and explained below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments provided in the present application without creative efforts fall within the scope of protection of the present application.

[0053] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application. For those of ordinary skill in the art, without creative efforts, the present application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in such a development process may be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing or production changes made based on the technical content disclosed in the present application are only conventional technical means and should not be understood as the content disclosed in the present application being insufficient.

[0054] Referring to "embodiments" in the present application means that the specific features, structures or characteristics described in connection with the embodiments may be included in at least one embodiment of the present application. The phrase appearing in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present application can be combined with other embodiments without conflict.

[0055] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the ordinary meanings understood by those with ordinary skills in the technical field to which this application pertains. The words such as "a", "an", "one kind", "the" and the like involved in this application do not indicate a limitation in quantity and may represent a singular or plural number. The terms "comprising", "including", "having" and any variations thereof involved in this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include steps or units not listed, or may further include other steps or units inherent to these processes, methods, products or devices. The words such as "connected", "linked", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The terms "first", "second", "third" and the like involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0056] An embodiment of this application provides a credit evaluation method for small and micro entities based on model fusion technology. Figure 1 It is a step flowchart of the credit evaluation method for small and micro entities based on model fusion technology according to the embodiment of this application, as Figure 1 shown, and this method includes the following steps:

[0057] Step S102, obtain training data and preprocess the training data to obtain standard data;

[0058] Specifically, since the collected training data has the characteristics of diversification and complexity and has different forms such as numbers and texts, first perform data preprocessing, convert it into standard data recognizable by a computer, and fill in the missing values according to business logic to facilitate subsequent separate binning thereof.

[0059] Step S104, perform binning on each index feature in the standard data to obtain several index bins of the index feature;

[0060] Specifically, Chi-Merge binning is performed on the standard data, that is, the variable values of each indicator feature in the standard data are sorted, each variable value is initialized as an indicator bin, and the chi-square values of adjacent two indicator bins are calculated traversally using the chi-square test. The two indicator bins with the smallest chi-square value are merged, and the traversal and merging are repeated until the preset number of indicator bins is obtained, or when the smallest chi-square value is greater than the preset threshold, the process stops.

[0061] It should be noted that using Chi-Merge binning to bin continuous features and some discrete features can ensure the stability of the subsequent model construction.

[0062] Step S106: Calculate the IV value of the indicator feature according to the indicator bin of the indicator feature, and screen the indicator feature based on the IV value to obtain the indicator feature with predictive ability.

[0063] Specifically, through the IV value calculation formula calculate the IV value of the indicator feature, where B i is the number of enterprises that default in the i-th indicator bin, B T is the number of all defaulting enterprises, G i is the number of enterprises that fulfill their obligations in the i-th indicator bin, G T is the number of all enterprises that fulfill their obligations;

[0064] The IV value represents the amount of information contained in the feature, that is, the influence degree of this feature on the model. Screen the indicator feature based on the IV value, and select the indicator feature with predictive ability whose IV value is greater than or equal to 0.02 (the indicator feature with an IV value less than 0.02 does not have predictive ability).

[0065] Step S108: Based on the stepwise regression principle and correlation test, screen out the second number of indicator features from the indicator features with predictive ability.

[0066] Specifically, the IV value in the above step S106 only represents the amount of information of a single feature and cannot explore the interaction between multi-indicator features. Therefore, based on the stepwise regression method to observe the effects of different feature combinations, the Akaike information criterion (AIC) is used as an indicator to screen out the first number of indicator features from the indicator features with predictive ability, which can achieve good fitting of the model to the data without overfitting.

[0067] The calculation formula of AIC is AIC = 2k - 2ln(L), where k is the number of indicator features and L is the likelihood function.

[0068] Specifically, through the correlation coefficient calculation formula Calculate the correlation coefficient between the first quantity of indicator features. Among them, X and Y are two different indicator features. If the correlation coefficient is greater than 0.5, then retain the indicator feature with the higher IV value among the two indicator features. Finally, obtain the second quantity of indicator features.

[0069] Step S110, calculate the weight of evidence for each bin of the second quantity of indicator features.

[0070] Specifically, through the encoding formula Calculate the weight of evidence for each bin of the second quantity of indicator features, where p i1 is the proportion of defaulting enterprises in the i-th bin of the indicator among all defaulting enterprises, p i0 is the proportion of performing enterprises in the i-th bin of the indicator among all performing enterprises, B i is the number of defaulting enterprises in the i-th bin of the indicator, B T is the number of all defaulting enterprises, G i is the number of performing enterprises in the i-th bin of the indicator, G T is the number of all performing enterprises.

[0071] It should be noted that WOE (weight of evidence) is a supervised encoding method that takes the attribute of the concentration of the predicted category as the encoded value, encodes the variable (feature), and converts the discrete variable into a continuous variable.

[0072] Step S112, based on the Logistic regression model, calculate the importance weights of the second quantity of indicator features.

[0073] Specifically, based on the mathematical representation of the Logistic regression model Calculate the importance weights of the second quantity of indicator features, where p is the probability of an enterprise defaulting, x is the indicator feature value, θ is the importance weight of the indicator feature, and T represents the transpose of the matrix.

[0074] Optionally, evaluate the model effect through indicators such as AUC and KS. AUC indicator: The area under the ROC curve, with a value range of [0, 1]. The closer it is to 1, the better the model prediction effect. The vertical axis of the ROC curve is TPR, and the horizontal axis is FPR; KS value: Measure the difference between the cumulative distributions of good and bad samples. The greater the cumulative difference between good and bad samples, the greater the KS indicator, that is, the stronger the risk discrimination ability of the model. KS = max|TPR - FPR|, that is, the maximum value of the absolute value of the difference between TPR and FPR. Among them, TPR is the ratio of the number of enterprises correctly judged as defaulting to the number of all defaulting enterprises. FPR is the ratio of the number of enterprises wrongly judged as defaulting to the number of all performing enterprises.

[0075] Step S114: Based on the weight of evidence and importance weight, construct the enterprise entity and enterprise owner scorecard models respectively.

[0076] Specifically, based on the weight of evidence and importance weight, construct a scorecard model; collect data from several data sources according to the unified social credit code, owner name, owner ID number, and owner mobile phone number of a preset enterprise. The data sources include the National Enterprise Credit Information Publicity System and the China Judgments Online, etc. The data types include industrial and commercial data, judicial data, intellectual property data, multi-headed data, equipment behavior data, consumption data, etc.; clean the data and convert it into a data type that meets the requirements of the scorecard model; input the converted data into the scorecard model for scoring, and the scoring reflects the performance probability of the preset enterprise in the credit scenario.

[0077] Preferably, construct a scorecard model: based on p as the probability of enterprise default and 1 - p as the probability of enterprise performance, transform the mathematical expression of the Logistic regression model into Define the relative probability of enterprise default as We can get ln(odds) = θ T x; perform score mapping through the formula Score = A - B * ln(Odds), where A = P0 - B * ln(Odds), PDO is the change value of the score when Odds doubles; P0 is the score when Odds remains unchanged;

[0078] Based on the above formula, combined with the importance weight of the index characteristics and the weight of evidence of each index bin of the index characteristics, we can get Score = A - B{θ i ω ij}, where θ i represents the importance weight of the i-th feature, and w ij represents the weight of evidence of the j-th index bin of the i-th index feature.

[0079] Optionally, further obtain the cut-off point through the scorecard model, that is, the optimal score for distinguishing enterprises on the verge of default from performing enterprises. First, bin the scores of the scorecard model, calculate the KS value of each bin, select the bin with the largest KS value as the candidate interval, and then make manual adjustments according to business requirements to select the best score as the cut-off.

[0080] Step S116: For the small and micro enterprise customer group and the self-employed industrial and commercial household customer group, fuse their enterprise entity scorecard models and enterprise owner scorecard models respectively to evaluate the credit level of small and micro enterprise customers in the credit scenario.

[0081] Specifically, through the weighted average algorithm, the prediction results of the enterprise entity scoring card model and the enterprise owner scoring card prediction results are calculated, the KS value of the fused model corresponding to each weight combination is calculated, and the optimal weight combination with the largest KS value is selected for the fusion of the enterprise entity and enterprise owner scoring card models to obtain the final fused model.

[0082] Through steps S102 to S116 in the embodiments of the present application, the problems of insufficient evaluation data, insufficient applicability of evaluation criteria, and insufficient number of modeling samples in the credit evaluation of small and micro enterprises in the credit scenario, resulting in poor model effects, are solved. Data is obtained from multiple sources, the credit evaluation criteria for small and micro enterprises are quantified, and the model is constructed and optimized through a large number of real samples to improve the accuracy of model prediction.

[0083] It should be noted that the steps shown in the above process or the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0084] The embodiments of the present application provide a credit evaluation system for small and micro entities based on model fusion technology. Figure 2 It is the structural block diagram of the credit evaluation system for small and micro enterprises in the credit scenario according to the embodiments of the present application. As Figure 2 shown, the system includes a data acquisition module 21, a feature screening module 22, a model construction module 23, and a model fusion module 24;

[0085] The data acquisition module 21 is used to acquire training data and preprocess the training data to obtain standard data;

[0086] The feature screening module 22 is used to bin each index feature in the standard data to obtain several index bins of the index feature; calculate the IV value of the index feature according to the index bins of the index feature, screen the index feature based on the IV value to obtain index features with predictive ability; based on the stepwise regression principle and correlation test, screen out a second number of index features from the index features with predictive ability;

[0087] The model construction module 23 is used to calculate the weight of evidence of each index bin in the second number of index features; calculate the importance weight value of the second number of index features based on the Logistic regression model; construct an enterprise entity and enterprise owner scoring card model based on the weight of evidence and the importance weight value;

[0088] The model fusion module 24 is used to calculate the optimal weight combination of the enterprise entity scoring card model and the enterprise owner scoring card model, obtain the finally fused model of the micro and small enterprise customer group and the self-employed industrial and commercial household customer group, and evaluate the credit level of enterprise customers in the credit scenario.

[0089] Through the data acquisition module 21, the feature screening module 22, the model construction module 23 and the model fusion module 24 in the embodiments of the present application, the problems of insufficient evaluation data, insufficient applicability of evaluation criteria, and insufficient number of modeling samples in the credit evaluation of micro and small enterprises in the credit scenario, resulting in poor model effects, are solved. It realizes the acquisition of data from multiple sources, quantifies the credit evaluation criteria of micro and small enterprises, and constructs and optimizes the model through a large number of real samples to improve the accuracy of model prediction.

[0090] Combined with the credit evaluation method of micro and small enterprises in the credit scenario in the above embodiments, the embodiments of the present application can be implemented by providing a storage medium. A computer program is stored on the storage medium; when the computer program is executed by a processor, any one of the credit evaluation methods of micro and small enterprises in the above embodiments is realized.

[0091] The processing device is as Figure 3 shown, and includes a processor, a memory, a communication interface and a bus. The processor, the memory and the communication interface are connected through the bus to complete the communication with each other. A computer program that can run on the processor is stored in the memory, and when the processor runs the computer program, it executes the micro-entity credit evaluation system model evaluation method in the above embodiments. The database is used to store the data during the execution process.

[0092] In some embodiments, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. In other embodiments, the processor may be a general-purpose processor of various types such as a central processing unit (CPU) and a digital signal processor (DSP), which is not limited here.

[0093] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0094] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A credit evaluation method for small and micro entities based on model fusion technology, characterized in that: The method comprises: Acquire training data, and preprocess the training data to obtain standard data; Binning each indicator feature in the standard data to obtain a plurality of indicator bins of the indicator feature; According to the indicator binning of the indicator feature, the IV value of the indicator feature is calculated, and the indicator feature is screened based on the IV value to obtain the indicator feature with predictive ability; Based on the stepwise regression principle and correlation test, a second number of indicator features are screened out from the indicator features with predictive capabilities; Calculating the weight of evidence for each indicator bin in the second number of indicator features; Based on the Logistic regression model, calculating the importance weights of the indicator features of the second quantity; Based on the evidence weights and the importance weights, constructing an enterprise subject scorecard model and an enterprise owner scorecard model; For the small and micro-enterprise customer groups and the individual business household customer groups, their corporate entity scoring card model and business owner scoring card model are respectively integrated to evaluate the credit level of small and micro-enterprises in credit scenarios.

2. The method according to claim 1, characterized in that Each indicator feature in the standard data is binned to obtain several indicator bins of the indicator feature, including: The variable values ​​of each indicator feature in the standard data are sorted, each variable value is initialized as an indicator bin, the chi-square values ​​of two adjacent indicator bins are traversed and calculated, the two indicator bins with the smallest chi-square values ​​are merged, and the traversal and merging are repeated until a preset number of indicator bins are obtained.

3. The method according to claim 1, characterized in that According to the indicator binning of the indicator feature, the IV value of the indicator feature is calculated, and the indicator feature is screened based on the IV value to obtain the indicator features with predictive ability, including: IV value calculation formula Calculate the IV value of the indicator feature, where B i is the number of enterprises that default in the i-th indicator bin, B T is the number of all defaulting enterprises, G i is the number of enterprises that fulfill their contracts in the i-th indicator bin, G T is the number of all performing enterprises; The indicator features are screened based on the IV value, and indicator features with predictive capabilities whose IV values ​​are greater than or equal to 0.02 are selected.

4. The method according to claim 1, characterized in that Based on the stepwise regression principle and correlation test, the second number of indicator features are screened out from the indicator features with predictive ability, including: Based on the stepwise regression principle, a first number of indicator features are screened out from the indicator features with predictive capabilities; Based on the correlation test, a second number of indicator features are screened out from the first number of indicator features.

5. The method according to claim 4, characterized in that: Based on the stepwise regression principle, a first number of indicator features are screened out from the indicator features with predictive capabilities, including: Akaike information criterion AIC=2k-2ln(L) is used as an indicator of the stepwise regression principle, and a first number of indicator features are screened out from the indicator features with predictive ability, wherein k is the number of indicator features and L is the likelihood function.

6. The method according to claim 4, characterized in that Screening out a second number of indicator features from the first number of indicator features based on the correlation test includes: The correlation coefficient is calculated by the formula The correlation coefficient between the first number of indicator features is calculated, where X and Y are two different indicator features. If the correlation coefficient is greater than 0.5, the indicator feature with a higher IV value among the two indicator features is retained, and finally the second number of indicator features is obtained.

7. The method according to claim 1, characterized in that Calculating the weight of evidence for each indicator bin in the second number of indicator features includes: By coding formula Calculate the weight of evidence for each indicator bin in the second number of indicator features, where p i1 is the proportion of defaulting enterprises in the ith indicator bin to all defaulting enterprises, p i0 is the proportion of enterprises that fulfill their contracts in the i-th indicator bin to all enterprises that fulfill their contracts, B i is the number of enterprises that default in the i-th indicator bin, B T is the number of all defaulting enterprises, G i is the number of enterprises that fulfill their contracts in the i-th indicator bin, G T is the number of all contract-compliant enterprises.

8. The method according to claim 1, characterized in that Based on the Logistic regression model, calculating the importance weight of the second number of indicator features includes: Mathematical representation based on Logistic regression model Calculate the importance weights of the indicator features of the second quantity, where p is the probability of enterprise default, x is the indicator feature value, θ is the importance weight of the indicator feature, and T represents the transpose of the matrix.

9. The method according to claim 1, characterized in that: Based on the evidence weight and the importance weight, constructing the enterprise subject scorecard model and the enterprise owner scorecard model includes: A scorecard model is constructed based on the evidence weights and the importance weights.

10. The method according to claim 1, characterized in that Based on the enterprise subject scorecard model and the enterprise owner scorecard model and the weighted average algorithm, the weight combination and model results of the final model fusion are calculated to evaluate the credit level of small and micro enterprises in credit scenarios, including: Based on the weighted average algorithm, the model results and KS value after the fusion of the enterprise subject scorecard model and the enterprise master scorecard model are calculated; The weight combination that maximizes the KS value of the fused model is selected as the optimal weight combination for the final model fusion; Based on the optimal weight combination, the final fusion model is constructed; Collect data from several data sources according to the company name of the preset enterprise, wherein the data sources include the National Enterprise Credit Information Formula System and the China Judgment Documents Network, etc., and the data types include industrial and commercial data, judicial data, intellectual property data, multi-head data, equipment behavior data, consumption data, etc.; Clean the data and convert it into a data type that meets the requirements of the scorecard model; Inputting the converted data into the scorecard model for scoring; The scoring of the scoring card model is integrated according to the optimal weight to obtain a final score, which reflects the probability of performance of the preset enterprise in the credit scenario.

11. A credit evaluation system for an enterprise in a credit scenario, characterized in that: The system includes a data acquisition module, a feature screening module, a model building module and a model fusion module; The data acquisition module is used to acquire training data and preprocess the training data to obtain standard data; The feature screening module is used to bin each indicator feature in the standard data to obtain a plurality of indicator bins of the indicator feature; calculate the IV value of the indicator feature according to the indicator bins of the indicator feature, and screen the indicator feature based on the IV value to obtain the indicator feature with predictive ability; based on the stepwise regression principle and the correlation test, screen a second number of indicator features from the indicator features with predictive ability; The model building module is used to calculate the evidence weight of each indicator bin in the second number of indicator features; Based on the Logistic regression model, calculating the importance weights of the indicator features of the second quantity; constructing an enterprise subject scorecard model and an enterprise owner scorecard model based on the evidence weights and the importance weights; The model fusion module is used to calculate the model result and KS value after the fusion of the enterprise subject scorecard model and the enterprise owner scorecard model, select the weight combination that maximizes the KS value, fuse the enterprise subject scorecard model and the enterprise owner scorecard model through a weighted average algorithm, and perform model fusion on the small and micro enterprise customer group and the individual business customer group respectively, to evaluate the credit level of small and micro enterprise customers in the credit scenario.

Images