Risk feature comprehensive evaluation method and device, electronic equipment and storage medium

Abstract

The invention discloses a risk feature comprehensive evaluation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a plurality of risk features in the retail credit field, and the types of the risk features comprise structured features and unstructured features; according to the types of the risk features, processing the risk features by a preset processing mechanism; performing data preprocessing on the plurality of risk features; determining an effectiveness score, a stability score, a correlation score and a business rationality score of each risk feature; determining a comprehensive score of the risk feature according to the utility score, the stability score, the correlation score and the business rationality score of the risk feature; and screening the risk features according to the comprehensive score of the plurality of risk features, and using the plurality of screened risk features for training of a preset model. The invention belongs to the field of risk monitoring. According to the method, risk features can be considered and screened from multiple dimensions.

Description

Technical Field

[0001] The present invention relates to the field of risk monitoring, and in particular to a comprehensive evaluation method, device, electronic equipment and storage medium for risk characteristics. Background Art

[0002] In the important field of retail credit data processing and analysis, the risk feature screening process plays a pivotal role. Specifically, this process aims to accurately and effectively identify the features that have the strongest influence on credit risk prediction from the complex and massive customer data, which is of great significance for effectively improving the prediction accuracy and overall operating efficiency of the credit risk prediction model.

[0003] Currently, the screening methods include manual screening and machine learning algorithm screening. Manual screening has the problem of low efficiency. When processing massive data, its performance is poor and it is difficult to meet the real-time requirements of rapid response. Although machine learning algorithm screening has the advantage of faster screening speed, it has obvious deficiencies in the in-depth understanding of the business and the interpretability of the results. Therefore, it is urgent to propose a new comprehensive evaluation method for risk characteristics. Summary of the invention

[0004] The present invention solves the technical problems of low risk feature screening efficiency and single dimension in the prior art by providing a comprehensive evaluation method, device, electronic device and storage medium for risk features, and achieves the technical effect of efficiently screening risk features from multiple dimensions.

[0005] In a first aspect, the present invention provides a comprehensive evaluation method for risk characteristics, the method comprising:

[0006] Acquire several risk features in the field of retail credit, wherein the types of risk features include structured features and unstructured features, and structured features include structured digital features and structured character features;

[0007] According to the type of risk characteristics, the risk characteristics are processed separately using the preset processing mechanism;

[0008] Data preprocessing for several risk characteristics;

[0009] Determine the utility score, stability score, relevance score and business rationality score of each risk feature; determine the comprehensive score of the risk feature based on the utility score, stability score, relevance score and business rationality score of the risk feature;

[0010] According to the comprehensive scores of several risk characteristics, the risk characteristics are screened, and several screened risk characteristics are used for training the preset model.

[0011] Furthermore, the utility score of the risk signature is determined, including:

[0012] U(x i ,y i )=w1*IV(x i )+w2*Gini(x i )+w3*FI(x i )+w4*MI(x i )

[0013] Among them, U is the utility score of the risk feature, x i is the i-th risk feature, y i is the fraud label of the ith risk feature, w1-w4 are preset weights, IV is the information value, Gini is the Gini coefficient, FI is the feature importance, and MI is the mutual information.

[0014] Furthermore, the stability score of the risk feature is determined, including:

[0015]

[0016] Among them, S is the stability score of the risk feature, x i is the ith risk feature, PSI(x i ) is the population stability index of the ith risk characteristic, PSI max It is the maximum value of the population stability index among several risk characteristics.

[0017] Further, determining the relevance score of the risk feature includes steps S131-S135:

[0018] Step S131, determining the correlation between the target risk feature and any other risk features;

[0019] Step S132, determining whether the correlation is lower than a preset threshold and whether the information value of the target risk feature is lower than the information value of any other risk feature;

[0020] Step S133: if the condition is satisfied, then the regularization penalty is applied to the target risk feature; if the condition is not satisfied, then the regularization penalty is not applied to the target risk feature;

[0021] Step S134, traverse each of the remaining risk features corresponding to the target risk feature, and loop through steps S131 to S133 to obtain a relevance score for the target risk feature.

[0022] Furthermore, regular penalties include:

[0023] Regularization(x i )=λ*IV(x i )

[0024] Among them, Regularization(x i ) is the result of the regularized penalty for the i-th risk feature, λ is the regularization coefficient, IV(x i ) is the information value of the ith risk feature.

[0025] Furthermore, according to the type of risk characteristics, the risk characteristics are processed respectively by the preset processing mechanism, including:

[0026] For structured character features, if the number of types of structured character features is less than 5, they are processed based on one-hot encoding;

[0027] If the number of types of structured character features is greater than or equal to 5, encoding processing is performed according to business characteristics.

[0028] Furthermore, data preprocessing is performed on several risk features, including:

[0029] Eliminate invalid or erroneous data based on business experience;

[0030] Delete the columns containing missing data and fill in the missing data based on interpolation, statistical methods or model prediction.

[0031] In a second aspect, the present invention provides a comprehensive evaluation device for risk characteristics, the device comprising:

[0032] An acquisition module, used for acquiring a number of risk features in the field of retail credit, wherein the types of risk features include structured features and unstructured features, and structured features include structured digital features and structured character features;

[0033] The first processing module is used to process the risk characteristics respectively according to the types of risk characteristics using a preset processing mechanism;

[0034] The second processing module is used to perform data preprocessing on a number of risk characteristics;

[0035] A scoring module is used to determine the utility score, stability score, relevance score and business rationality score of each risk feature; a comprehensive evaluation module is used to determine the comprehensive score of the risk feature based on the utility score, stability score, relevance score and business rationality score of the risk feature;

[0036] The training module is used to screen risk features according to the comprehensive scores of several risk features, and use the screened risk features for training a preset model.

[0037] In a third aspect, the present invention provides an electronic device, comprising:

[0038] processor;

[0039] a memory for storing processor-executable instructions;

[0040] Wherein, the processor is configured to execute to implement a comprehensive evaluation method of risk characteristics as provided in the first aspect.

[0041] In a fourth aspect, the present invention provides a non-temporary computer-readable storage medium, which, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute a comprehensive evaluation method for risk characteristics as provided in the first aspect.

[0042] One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

[0043] The present invention provides a comprehensive evaluation method for risk characteristics, the method comprising: obtaining several risk characteristics in the field of retail credit, wherein the types of risk characteristics include structured characteristics and unstructured characteristics, and the structured characteristics include structured digital characteristics and structured character characteristics; processing the risk characteristics respectively according to the types of risk characteristics with a preset processing mechanism; performing data preprocessing on several risk characteristics; determining the utility score, stability score, relevance score and business rationality score of each risk characteristic; determining the comprehensive score of the risk characteristic according to the utility score, stability score, relevance score and business rationality score of the risk characteristic; screening the risk characteristics according to the comprehensive scores of several risk characteristics, and using the several screened risk characteristics for training a preset model. The present invention constructs a comprehensive and efficient feature set, introduces multi-attribute utility theory (MAUT), and comprehensively evaluates the utility, stability, relevance and business rationality of risk characteristics. Then, a comprehensive evaluation scoring function of multi-dimensional risk characteristics is constructed, and all characteristics are scored and sorted. The present invention enhances the speed and automation of the risk characteristic screening process by reasonably setting the screening means, and improves the work efficiency. The present invention introduces the Multi-Attribute Utility Theory (MAUT), which improves the interpretability of the screening results, improves the comprehensiveness and accuracy of feature screening, and can better meet actual business needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 A schematic diagram of a flow chart of a comprehensive evaluation method for risk characteristics provided by the present invention. DETAILED DESCRIPTION

[0046] The embodiment of the present invention solves the technical problems of low efficiency and single dimension in risk feature screening in the prior art by providing a comprehensive evaluation method for risk features.

[0047] The technical solution of the present invention is to solve the above technical problems, and the overall idea is as follows:

[0048] A comprehensive evaluation method for risk characteristics, the method comprising: obtaining a number of risk characteristics in the field of retail credit, wherein the types of risk characteristics include structured characteristics and unstructured characteristics, and the structured characteristics include structured digital characteristics and structured character characteristics; processing the risk characteristics respectively according to the types of risk characteristics with a preset processing mechanism; performing data preprocessing on a number of risk characteristics; determining the utility score, stability score, relevance score and business rationality score of each risk characteristic; determining a comprehensive score of the risk characteristic according to the utility score, stability score, relevance score and business rationality score of the risk characteristic; screening the risk characteristics according to the comprehensive scores of the risk characteristics, and using the screened risk characteristics for training a preset model.

[0049] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0050] First of all, the term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. 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, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0051] In the important field of retail credit data processing and analysis, the screening of risk characteristics is an important step. This step aims to accurately and effectively identify those characteristics that have the strongest influence on credit risk prediction from the complex and massive customer data. The screened risk characteristics are conducive to improving the prediction accuracy of the credit risk prediction model.

[0052] The present invention provides Figure 1 A comprehensive evaluation method for risk characteristics shown includes steps S11-S15:

[0053] Step S11, obtaining a number of risk features in the field of retail credit, wherein the types of risk features include structured features and unstructured features, and structured features include structured digital features and structured character features.

[0054] In the field of retail credit data processing and analysis, risk characteristics are variables or attributes that can affect credit risk. Risk characteristics can be obtained directly from information provided by customers (such as age, income, occupation), or derived from transaction behavior, historical records and other data (such as credit score, payment habits).

[0055] Structured features refer to data that can be directly stored in a relational database. Structured features usually have a fixed format and length. In the present invention, structured features include but are not limited to basic user information (such as age, gender, occupation, etc.), device information (such as device model, operating system version, etc.), embedded point information (such as user click, browse and other behavior records), etc.

[0056] Structured features include structured digital features and structured character features. Unstructured features refer to data without a fixed format and length, such as image information, filled-in text information, voice information, behavior time sequence information, and knowledge graph features.

[0057] Each risk signature contains several pieces of data.

[0058] Step S12: Process the risk features respectively according to the types of risk features using a preset processing mechanism.

[0059]

Structural features

[0060] {Structured digital features}: For structured digital features, the structured digital features themselves are already in numerical form and can be directly used for subsequent analysis and processing.

[0061] {Structured character features}: Since they are in text form, they need to be converted into numerical form for analysis. Specifically, for structured character features, if the number of types of structured character features is less than 5, they are processed based on one-hot encoding; if the number of types of structured character features is greater than or equal to 5, they are encoded according to business characteristics.

[0062] One-hot encoding is a method of converting categorical variables into a binary matrix where each categorical variable has a corresponding column that is 1 if the value of the categorical variable appears in the sample and 0 otherwise.

[0063] If the number of character features is greater than or equal to 5, they need to be encoded according to business characteristics. For example, for provincial information, you can sort them according to the GDP of the province, and then use the sorted index as the encoding of the feature.

[0064] Unstructured features

[0065] Unstructured features can be processed in different ways according to the fields in which they are located. For example, in the field of image data, in addition to the traditional CNN method, Vision Transformer (ViT) is also used to process image block sequences and convert image data into image vector features. For text data, the Transformer architecture model BERT is used to deeply understand the text semantics and contextual information through the self-attention mechanism and convert text data into text vector features. For time series data, Transformer is also used to model long-term dependencies in time series data and convert time series data into time series vector features. For page buried data, by combining the sequence modeling capability of Transformer, the page buried data is combined with the time series dimension and converted into time series vector features of operation behavior.

[0066] Through steps S11-S12, the risk features after processing can be saved in the database for subsequent risk assessment and analysis. Aggregation operations can be performed on the above risk features to integrate the scattered risk feature information, thereby mining more in-depth and insightful aggregate features.

[0067] For example, by aggregating the same account manager, we can get aggregated risk characteristics. In addition to the average age of customers, we can also calculate statistics such as the median age and age standard deviation of the customers under the account manager. The above risk characteristics can more comprehensively reflect the age distribution of the customer group that the account manager is responsible for. In addition, other types of aggregation operations can be performed. For example, we can calculate aggregated risk characteristics such as the average asset size of customers under the same GPS to help better understand the situation in different regions, or use different regions as aggregation units to calculate the average income level of the region.

[0068] Step S13, performing data preprocessing on a number of risk features.

[0069] The main purpose of data preprocessing is to ensure the quality and accuracy of data and to facilitate data cleaning of converted features. Specifically, it includes removing invalid or erroneous risk feature data and filling in missing risk feature data. Based on business experience and data requirements, it detects whether the risk feature data conforms to the predefined pattern.

[0070] Specifically include: eliminating invalid or erroneous data based on business experience; deleting columns or rows containing missing data, and filling in missing data based on interpolation, statistical methods or model predictions.

[0071] Based on business experience, identify and eliminate data segments with risk characteristics of historical data calculation errors caused by system reasons.

[0072] The types of missing values ​​include row missing values ​​and column missing values. Row missing values ​​and column missing values ​​are two different types of missing data. Row missing values ​​refer to missing data in some rows of a data set, while column missing values ​​refer to missing data in some columns of a data set.

[0073] When there are missing rows, you can delete the rows with missing values, use statistical methods or model predictions to fill in the missing values, or use interpolation methods to estimate the missing values.

[0074] When there are missing rows, you can delete the columns containing missing values, use statistical methods or model predictions to fill the missing values, use interpolation methods to estimate the missing values, or use a classification model to predict the missing values.

[0075] For structured digital features, the present invention also provides a filling method, including:

[0076] Calculate the missing value ratio of the row, including: missing value ratio = number of missing characteristic risks / total number of characteristic risks. Eliminate rows with missing value ratios higher than the preset threshold. For the eliminated characteristic risks, the missing values ​​of the missing columns can be filled, which can be filled based on the mean of the structured digital features.

[0077] Mean of structured numeric features, including:

[0078]

[0079] Where P is the mean, n is the number of non-empty structured digital features, and x i is the ith risk feature.

[0080] The filling method depends on the type of risk feature, the type of gap, and business needs. For example, if the data is unevenly distributed, using the mean to fill may lead to bias, and the median or mode filling may be more appropriate. For time series data, interpolation filling usually better preserves the temporal characteristics of the data. The cleaned data will be used for subsequent feature screening to ensure that the analysis can be based on high-quality data.

[0081] In addition, to further improve the data quality of risk features, pre-defined validation rules can be applied to detect whether the data conforms to pre-defined patterns based on business experience and data requirements. For example, regular expressions can be used to verify the date format or phone number format. In addition, custom pattern recognition algorithms can be used to identify specific patterns in the data based on business logic, identify duplicate records, detect inconsistencies in the data, etc.

[0082] In addition, data visualization tools can be used to create charts and graphs to obtain data distribution, trends, and anomalies, and delete data with abnormal risk characteristics.

[0083] Step S14, determining the utility score, stability score, relevance score and business rationality score of each risk feature; determining the comprehensive score of the risk feature based on the utility score, stability score, relevance score and business rationality score of the risk feature.

[0084] Determine the utility score of the risk signature, including:

[0085] U(x i ,y i )=w1*IV(x i )+w2*Gini(x i )+w3*FI(x i )+w4*MI(x i )

[0086] Among them, U is the utility score of the risk feature, x i is the i-th risk feature, y i is the fraud label of the ith risk feature, w1-w4 are preset weights, IV is the information value, Gini is the Gini coefficient, FI is the feature importance, and MI is the mutual information.

[0087] The utility function U is used to evaluate the effectiveness of risk feature X. Feature set X is a data set containing multiple types of features. The mathematical representation of feature set X is (X = x1, x2, ..., x n ), where x i represents i risk characteristics.

[0088] The feature set Y is used to define labels to identify whether the risk feature hits a certain fraud scenario. The mathematical representation of the feature set Y is Y = y1, y2, …, y m ,y i is the fraud label of the i-th risk feature.

[0089] Determine the stability score of the risk signature, including:

[0090]

[0091] Among them, S(x i ) is the risk characteristic x i The stability score, x i is the ith risk feature, PSI(x i ) is the population stability index of the ith risk characteristic, PSI max It is the maximum value of the population stability index among several risk characteristics.

[0092] The stability utility function S is used to make an in-depth and accurate assessment of the stability of a feature, specifically including: a pre-set time window, and calculating the Population Stability Index (PSI) in each different window.

[0093] The selection of time windows is very flexible, and different months can be selected according to actual needs. By pre-setting the time window, the stability performance of the feature in different time periods can be analyzed in an all-round way, providing a more accurate and reliable basis for subsequent risk assessment and management. By analyzing the time windows of different months, the trend of feature stability changing over time can be better captured.

[0094] Determining the relevance score of the risk feature includes steps S131-S135:

[0095] Step S131, determining the correlation between the target risk feature and any other risk features, specifically includes:

[0096]

[0097] Among them, x i is the ith risk feature, x k is the kth risk feature, and m is the number of bins (x i It includes several data, and the data can be binned). i is the proportion of good samples in the i-th bin, q i is the proportion of bad samples in the i-th bin.

[0098] Step S132, determining whether the correlation is lower than a preset threshold and whether the information value of the target risk feature is lower than the information value of any other risk feature.

[0099] Information value is an indicator used to measure the predictive ability of risk characteristics on target variables (such as risk status in risk assessment). Information value is widely used in many fields such as data mining and credit risk assessment.

[0100] Determining the value of information includes:

[0101]

[0102] Where IV is the information value, O is the number of groups (the k-th risk feature includes several data, and the data can be grouped). is the bad sample rate, For a good sample rate.

[0103] Step S133: if the condition is satisfied, then the regularization penalty is applied to the target risk feature; if the condition is not satisfied, then the regularization penalty is not applied to the target risk feature;

[0104] Regularization penalties include:

[0105] Regularization(x i )=λ*IV(x i )

[0106] Among them, Regularization(x i ) is the result of the regularized penalty for the i-th risk feature, λ is the regularization coefficient, IV(x i ) is the information value of the ith risk feature.

[0107] Step S134, traverse each of the remaining risk features corresponding to the target risk feature, and execute steps S131 to S133 in a loop to obtain the relevance score of the target risk feature, that is, C(x i , x n-1 ), where n is the number of risk features.

[0108] Traverse each of the remaining risk features corresponding to the target risk feature, obtain several results of applying regularized penalties to the target risk feature, and use the several results of applying regularized penalties to the target risk feature as the correlation score corresponding to the target risk feature.

[0109] We can define the business rationality utility function BR(x i ) Evaluate risk characteristics x i Business rationality assessment can be performed by combining domain knowledge and business rules. For example, for a certain feature, if its vacancy rate is too high or the value fluctuates too much, the business reasons behind it can be further investigated.

[0110] Step S15, screening the risk features according to the comprehensive scores of the risk features, and using the screened risk features for training the preset model.

[0111] L i =U(x i ,y i )*S(x i )*C(x i , x n-1 )*BR(x i )

[0112] Among them, L i is the comprehensive score of the ith risk feature.

[0113] Specifically, the top G comprehensive scores can be screened out, and the risk features corresponding to the top G comprehensive scores can be applied to the training of the preset model.

[0114] Specifically, after retaining the first G risk features, the first G risk features can be further preprocessed, such as normalization and standardization, to make the risk features more reasonable in numerical range and distribution, which is conducive to model training and optimization.

[0115] Select appropriate model algorithms, such as decision trees, random forests, support vector machines, etc., and use the retained risk features to train the model. During the training process, methods such as cross-validation can be used to evaluate the performance of the model and adjust the model parameters in a timely manner to improve the accuracy and generalization ability of the model. At the same time, pay close attention to the training progress and resource consumption of the model to ensure that the model training is completed within a reasonable time and resource range.

[0116] In summary, the present invention provides a comprehensive evaluation method for risk characteristics, the method comprising: obtaining several risk characteristics in the field of retail credit, wherein the types of risk characteristics include structured characteristics and unstructured characteristics, and structured characteristics include structured digital characteristics and structured character characteristics; processing the risk characteristics respectively according to the types of risk characteristics with a preset processing mechanism; performing data preprocessing on several risk characteristics; determining the utility score, stability score, relevance score and business rationality score of each risk characteristic; determining the comprehensive score of the risk characteristic according to the utility score, stability score, relevance score and business rationality score of the risk characteristic; screening the risk characteristics according to the comprehensive scores of several risk characteristics, and using the several screened risk characteristics for training a preset model. The present invention constructs a comprehensive and efficient feature set, introduces multi-attribute utility theory (MAUT), and comprehensively evaluates the utility, stability, relevance and business rationality of risk characteristics. Then, a comprehensive evaluation scoring function of multi-dimensional risk characteristics is constructed, and all characteristics are scored and sorted. The present invention enhances the speed and automation of the risk characteristic screening process by reasonably setting the screening means, and improves work efficiency. The present invention introduces the Multi-Attribute Utility Theory (MAUT), which improves the interpretability of the screening results, improves the comprehensiveness and accuracy of feature screening, and can better meet actual business needs.

[0117] Based on the same inventive concept, the present invention provides a comprehensive evaluation device for risk characteristics, the device comprising:

[0118] An acquisition module, used for acquiring a number of risk features in the field of retail credit, wherein the types of risk features include structured features and unstructured features, and structured features include structured digital features and structured character features;

[0119] The first processing module is used to process the risk characteristics according to the types of risk characteristics using a preset processing mechanism;

[0120] The second processing module is used to perform data preprocessing on a number of risk characteristics;

[0121] A scoring module is used to determine the utility score, stability score, relevance score and business rationality score of each risk feature; a comprehensive evaluation module is used to determine the comprehensive score of the risk feature based on the utility score, stability score, relevance score and business rationality score of the risk feature;

[0122] The training module is used to screen risk features according to the comprehensive scores of several risk features, and use the screened risk features for training a preset model.

[0123] Based on the same inventive concept, the present invention also provides an electronic device, including:

[0124] processor;

[0125] a memory for storing processor-executable instructions;

[0126] The processor is configured to execute to implement a comprehensive evaluation method of risk characteristics as provided above.

[0127] Based on the same inventive concept, the present invention also provides a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute a comprehensive evaluation method for risk characteristics as provided above.

[0128] Since the electronic device introduced in this embodiment is an electronic device used to implement the information processing method in the embodiment of the present invention, based on the information processing method introduced in the embodiment of the present invention, a person skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present invention is not described in detail here. As long as the electronic device used by a person skilled in the art to implement the information processing method in the embodiment of the present invention, it belongs to the scope of protection of the present invention.

[0129] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0133] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0134] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A comprehensive evaluation method for risk characteristics, characterized in that: The method comprises: Acquire several risk features in the field of retail credit, wherein the types of risk features include structured features and unstructured features, and structured features include structured digital features and structured character features; According to the type of risk characteristics, the risk characteristics are processed separately using the preset processing mechanism; Data preprocessing for several risk characteristics; Determine the utility score, stability score, relevance score, and business plausibility score for each risk feature; Determine the comprehensive score of the risk feature based on the risk feature's utility score, stability score, relevance score, and business rationality score; According to the comprehensive scores of several risk characteristics, the risk characteristics are screened, and several screened risk characteristics are used for training the preset model.

2. A comprehensive evaluation method for risk characteristics according to claim 1, characterized in that: Determine the utility score of the risk signature, including: U(x i y i )=w1*IV(x i )+w2*Gin(x i )+w3*FI(x i )+w4*MI(x i ) Among them, U is the utility score of the risk feature, x i is the i-th risk feature, y i is the fraud label of the ith risk feature, w1-w4 are preset weights, IV is the information value, Gini is the Gini coefficient, FI is the feature importance, and MI is the mutual information.

3. A comprehensive evaluation method for risk characteristics according to claim 1, characterized in that: Determine the stability score of the risk signature, including: Among them, S is the stability score of the risk feature, x i is the ith risk feature, PSI(x i ) is the population stability index of the ith risk characteristic, PSI max It is the maximum value of the population stability index among several risk characteristics.

4. A comprehensive evaluation method for risk characteristics according to claim 1, characterized in that: Determining the relevance score of the risk feature includes steps S131-S135: Step S131, determining the correlation between the target risk feature and any other risk features; Step S132, determining whether the correlation is lower than a preset threshold and whether the information value of the target risk feature is lower than the information value of any other risk feature; Step S133: if the condition is satisfied, then the regularization penalty is applied to the target risk feature; if the condition is not satisfied, then the regularization penalty is not applied to the target risk feature; Step S134, traverse each of the remaining risk features corresponding to the target risk feature, and execute steps S131 to S133 in a loop to obtain a correlation score of the target risk feature.

5. A comprehensive evaluation method for risk characteristics according to claim 1, characterized in that: Regularization penalties include: Regularization(x i )=λ*IV(x i ) Among them, Regularization(x i ) is the result of the regularized penalty for the i-th risk feature, λ is the regularization coefficient, IV(x i ) is the information value of the ith risk feature.

6. A comprehensive evaluation method for risk characteristics according to claim 1, characterized in that: According to the type of risk characteristics, the risk characteristics are processed separately using the preset processing mechanism, including: For structured character features, if the number of types of structured character features is less than 5, they are processed based on one-hot encoding; If the number of types of structured character features is greater than or equal to 5, encoding processing is performed according to business characteristics.

7. A comprehensive evaluation method for risk characteristics according to claim 1, characterized in that: Data preprocessing for several risk features, including: Eliminate invalid or erroneous data based on business experience; Delete the columns or rows containing missing data and fill in the missing data based on interpolation, statistical methods or model prediction.

8. A comprehensive evaluation device for risk characteristics, characterized in that: The device comprises: An acquisition module, used for acquiring a number of risk features in the field of retail credit, wherein the types of risk features include structured features and unstructured features, and structured features include structured digital features and structured character features; The first processing module is used to process the risk characteristics respectively according to the types of risk characteristics using a preset processing mechanism; The second processing module is used to perform data preprocessing on a number of risk characteristics; A scoring module is used to determine the utility score, stability score, relevance score and business rationality score of each risk feature; a comprehensive evaluation module is used to determine the comprehensive score of the risk feature based on the utility score, stability score, relevance score and business rationality score of the risk feature; The training module is used to screen risk features according to the comprehensive scores of several risk features, and use the screened risk features for training a preset model.

9. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; Wherein, the processor is configured to execute to implement a comprehensive evaluation method of risk characteristics as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to implement a comprehensive evaluation method for risk characteristics as described in any one of claims 1 to 7.

Images