Personal credit assessment method and system based on dual cost sensitive loss function, and server

By introducing a double cost sensitive loss function and a fuzzy C-mean algorithm into the credit scoring model, combined with the taboo search algorithm, the sample imbalance problem is solved, and the accuracy of the credit evaluation model and the ability to identify difficult-to-separate customer samples are improved.

CN118982422AInactive Publication Date: 2024-11-19ZHEJIANG FULIN TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411477447.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-22
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

When dealing with customer sample imbalance, the existing credit scoring model is susceptible to initial point setting and abnormal data, making it difficult to accurately evaluate and classify customer samples, resulting in a decrease in evaluation accuracy.

Method used

Using a personal credit evaluation method based on double-cost sensitive loss function, samples are classified through fuzzy C-mean algorithm and taboo search algorithm, and equilibrium factor is introduced in the distributed gradient enhancement decision tree XGBoost, and the loss function is adjusted to optimize the model for a few class samples. recognition ability.

Benefits of technology

It significantly improves the accuracy of the personal credit evaluation model, reduces the interference of initial point setting and abnormal data on the model, improves the ability to identify a few types of samples, and enhances the reliability of the evaluation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982422A_ABST
    Figure CN118982422A_ABST
Patent Text Reader

Abstract

The invention discloses a personal credit assessment method and system based on a dual cost sensitive loss function, and a server, and the method comprises the steps: carrying out the preprocessing of data, carrying out the matching of corresponding features, marking a label for each sample through the rolling rate analysis and account age analysis, on the basis, a taboo search algorithm and a fuzzy C-means algorithm are combined to construct category features to classify similar crowd samples; then introducing a balance factor alpha to optimize a focusing loss function to construct a dual cost sensitive loss function, and adjusting an optimization target of the tree model to train the focusing loss function; then, using the preprocessed data, adding classification features, and adjusting an optimization objective function of a distributed gradient boosting decision tree XGBoost to be a dual cost sensitive loss function to train a personal credit assessment model; and finally, inputting the to-be-evaluated personal information data into the trained personal credit evaluation model to obtain a corresponding credit evaluation result, thereby reducing the interference of sample imbalance on the model, and improving the model evaluation accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a personal credit evaluation method, system and server based on a dual cost-sensitive loss function. Background Art

[0002] The credit scoring model is a key quantitative tool for measuring the credit status of customers. As the core reference for the final credit approval, it directly determines the customer's credit limit and interest rate level, and is crucial to ensuring the safety of repayment business. However, the risk scoring model construction methods commonly used in the market currently face the problems of large differences in customer groups and imbalanced positive and negative samples. For example, there are large differences in customer income between samples and a very low proportion of positive samples. This situation will interfere with model judgment and reduce accuracy. For example, in the model training stage, customer samples can be divided into easy-to-classify customer samples and difficult-to-classify customer samples. There are often a large number of easy-to-classify customer samples and a small number of difficult-to-classify customer samples in the training set. The prediction of difficult-to-classify samples is more difficult and more important for the model. The loss of easy-to-classify samples is very small, but due to their large number, there is still a possibility of dominating the learning direction of the model. The traditional model combined with the cost-sensitive factor does not produce more bias for difficult-to-classify customer samples. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a personal credit assessment method based on a dual cost-sensitive loss function, comprising the following steps: S1, preprocessing the data, including obtaining the personal information data and loan behavior time of each sample, matching the corresponding features, and labeling each sample through rolling rate analysis and aging analysis. The features include but are not limited to historical repayment behavior and historical credit records of each institution. The personal information data includes ID card, mobile phone number and name; S2, combining the taboo search algorithm and the fuzzy C-means algorithm to construct category features to classify similar population samples; S3, construct a dual cost-sensitive loss function, introduce a balance factor α to optimize the focus loss function, and adjust the optimization target of the tree model to the focus loss function for training; S4, using the preprocessed data, adding classification features, and adjusting the optimization objective function of the distributed gradient boosting decision tree XGBoost to a dual cost-sensitive loss function to train the personal credit assessment model; S5, inputting the personal information data to be evaluated into the trained personal credit evaluation model to obtain the corresponding credit evaluation result.

[0004] Preferably, the step S2 specifically includes: S21, obtain a set of initial cluster centers through the fuzzy C-means algorithm for data clustering operations in unsupervised learning , remember the current point , global optimal point , calculate the optimal fitness function value , leave the taboo table empty; S22, determine whether the stopping criterion is met, the stopping criterion is configured to use the maximum number of iterations criterion, that is, iterate until the set number of times stops, if it is met, terminate the algorithm process, otherwise, Randomly pick points in the neighborhood Second, based on the collected points, the fuzzy C-means algorithm is used to perform local optimization and generate Neighborhood solution sequence ; S23, the obtained neighborhood solution sequence , calculate the fitness function values ​​of all candidate solutions , and get the optimal solution And the corresponding adaptation value ,judge Whether the amnesty rules are met, if the amnesty rules are met, Higher than , update the global optimal point , the optimal fitness function value ; If the amnesty rules are not met, then judge Is it in a taboo state? If not, remember the current point. ; If it is in a taboo state, select the best solution from the solutions in the non-taboo state and record it as ; If all points are in the taboo state, take the linear average of all points in the taboo table as the new current solution ; S24, will Add the item to the taboo list and update the taboo list according to the length of the taboo list, and return to step S22; S25, after the stopping criterion is met, the samples are classified based on the final cluster centers.

[0005] Preferably, the step S21 specifically includes: S211, set the number of cluster centroids c and fuzzy index , initialize the centroids of each cluster , set the convergence accuracy , let the number of iterations n=0; S212, calculation sample To cluster centroid The Euclidean distance , update membership ;like < , then let ; Get a new set of membership ; S213, obtain a new set of cluster centroids based on the new membership If calculated , then the algorithm stops; otherwise, And jump to step S212 to continue execution.

[0006] Preferably, the loss function is: ,in Focus on adjusting the weights of minority categories, Focus on adjusting the weights of misclassified samples. The weight selection is configured to be based on the Bayesian parameter adjustment strategy or the grid parameter adjustment strategy. Output the corresponding sample probability for the model, .

[0007] The present invention also discloses a personal credit evaluation system based on a dual cost-sensitive loss function, comprising a sample processing module, a sample classification module, a function optimization module, a model training module and a result acquisition module, wherein the sample processing module is used to pre-process data, including obtaining various personal information data and loan behavior time as samples, matching corresponding features, and labeling each sample through rolling rate analysis and account age analysis, wherein the features include but are not limited to historical repayment behaviors and historical credit records of various institutions, and the personal information data includes identity cards, mobile phone numbers and names; the sample classification module is used to combine a taboo search algorithm and a fuzzy C-means algorithm to construct category features to classify similar population samples; the function optimization module is used to construct a dual cost-sensitive loss function, introduce a balance factor α to optimize the focus loss function, and adjust the optimization target of the tree model to the focus loss function for training; the model training module is used to use the pre-processed data, add classification features, and adjust the optimization target function of the distributed gradient boosting decision tree XGBoost to the dual cost-sensitive loss function to train the personal credit evaluation model; the result acquisition module is used to obtain the corresponding credit evaluation result after inputting the personal information data to be evaluated into the trained personal credit evaluation model.

[0008] Preferably, the sample classification module specifically includes an initial configuration module, a local optimization module, an optimal solution module, a taboo update module and a sample classification module, wherein the initial configuration module is used to obtain a set of initial cluster centers by using a fuzzy C-means algorithm for data clustering operations in unsupervised learning. , remember the current point , global optimal point , calculate the optimal fitness function value , set the taboo table to empty; local optimization module, used to determine whether the stopping criterion is met, the stopping criterion is configured to use the maximum number of iterations criterion, that is, iterate to the set number of times to stop, if it is satisfied, the algorithm process is terminated, otherwise the current point Randomly pick points in the neighborhood Second, based on the collected points, the fuzzy C-means algorithm is used to perform local optimization and generate Neighborhood solution sequence ; Optimal solution module, used to obtain the neighborhood solution sequence , calculate the fitness function values ​​of all candidate solutions , and get the optimal solution And the corresponding adaptation value ,judge Whether the amnesty rules are met, if the amnesty rules are met, Higher than , update the global optimal point , the optimal fitness function value ; If the amnesty rules are not met, then judge Is it in a taboo state? If not, remember the current point. ; If it is in a taboo state, select the best solution from the solutions in the non-taboo state and record it as ; If all points are in the taboo state, take the linear average of all points in the taboo table as the new current solution ; Taboo update module, used to Add to the taboo table and update the taboo table according to the length of the taboo table, and return to the local optimization module; the sample classification module is used to classify the samples based on the final cluster center after meeting the stopping criteria.

[0009] Preferably, the initial configuration module further includes a parameter configuration module, a membership updating module and a cluster centroid updating module, wherein the parameter configuration module is used to set the number of cluster centroids c and the fuzzy index , initialize the centroids of each cluster , set the convergence accuracy , record the number of iterations n=0; membership update module, used to calculate the sample To cluster centroid The Euclidean distance , update membership ;like < , then let ; Get a new set of membership ; Cluster centroid update module, used to obtain a new set of cluster centroids based on the new membership If calculated , then the algorithm stops; otherwise, And jump to step S212 to continue execution.

[0010] Preferably, the loss function is: ,in Focus on adjusting the weights of minority categories, Focus on adjusting the weights of misclassified samples. The weight selection is configured to be based on the Bayesian parameter adjustment strategy or the grid parameter adjustment strategy. Output the corresponding sample probability for the model, .

[0011] The present invention also discloses a server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the aforementioned methods when executing the computer program.

[0012] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any of the aforementioned methods are implemented.

[0013] The personal credit assessment method, system and server based on the dual cost-sensitive loss function disclosed in the present invention, after pre-processing and cleaning the training sample data, classify similar sample populations by introducing a taboo search algorithm in a fuzzy C-means algorithm, thereby effectively improving the clustering quality and discrimination of the sample data, and at the same time constructing a dual cost-sensitive loss function by introducing a balance factor in a focus loss function, and adjusting the optimization objective function of the distributed gradient boosting decision tree XGBoost to the dual cost-sensitive loss function to consider the cost difference between different samples, effectively dealing with the problem of sample imbalanced data sets, and improving the model's recognition ability for minority class samples or high-cost errors, and finally inputting the sample data into a personal credit assessment model for training, so that the model can obtain corresponding credit assessment results according to the input personal information data, significantly improving the accuracy of the personal credit assessment model, and reducing the interference of the initial point setting, abnormal data, and sample imbalance on the model.

[0014] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 The figure is a schematic diagram of the steps of a personal credit assessment method based on a dual cost-sensitive loss function disclosed in one embodiment of the present invention.

[0016] Figure 2 It is a schematic diagram of the structure of a personal credit evaluation system based on a dual cost-sensitive loss function disclosed in another embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0019] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0020] Unless otherwise defined, the technical or scientific terms used herein shall have the common meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the patent application specification and claims of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "a" do not indicate a quantity limitation, but indicate the existence of at least one.

[0021] The credit scoring model is a key quantitative tool for measuring a customer's credit status. As the core reference for the final credit approval, it directly determines the customer's credit limit and interest rate level, and is crucial to ensuring the safety of repayment business. However, the risk scoring model construction strategy widely used in the current market mostly relies on clustering algorithms and cost-sensitive learning risk control models. Although these methods are effective, they are easily affected by the initial point setting and abnormal data. In addition, the significant differences between customer groups lead to imbalanced model training sample data, which in turn affects the judgment accuracy of the model and reduces the evaluation accuracy.

[0022] Therefore, in this embodiment, a personal credit assessment method based on a dual cost-sensitive loss function is disclosed, as shown in the attached Figure 1 As shown, the following steps are included.

[0023] Step S1, preprocessing the data, including obtaining various personal information data and loan behavior time as samples, matching corresponding features, and labeling each sample through rolling rate analysis and aging analysis. The features include but are not limited to historical repayment behavior and historical credit records of various institutions. The personal information data includes ID card, mobile phone number and name.

[0024] Specifically, we first preprocess the sample data and organize the three elements of the samples obtained: ID card, mobile phone number and name, and record the submission time of each sample. Then, we match these basic characteristics with a series of characteristics including historical repayment behavior and historical credit records of various other institutions to comprehensively evaluate the credit status of each sample. Finally, we use the rolling rate analysis and aging analysis methods to label each sample with a y label to judge its quality, and comprehensively and scientifically evaluate the credit status of each sample, providing strong support for the credit assessment decisions of subsequent models.

[0025] Step S2, combining the taboo search algorithm and the fuzzy C-means algorithm to construct category features to classify similar population samples.

[0026] Specifically, the fuzzy C-means algorithm is a clustering algorithm based on fuzzy set theory. It allows data points to belong to multiple clusters at the same time, and uses a membership function to represent the degree of membership of each data point to each cluster. It is usually used for data clustering problems in unsupervised learning. The taboo search algorithm (TS) is a meta-heuristic random search algorithm that aims to find the global optimal solution by avoiding repeated searches and falling into local optimal solutions. During the implementation process, it starts from an initial feasible solution, selects a series of specific search direction movements as a trial, and selects the movement that makes the specific objective function value change the most. In order to avoid falling into the local optimal solution, the TS search records and selects the optimization process that has been carried out by establishing a taboo table to guide the next search direction. This embodiment uses a combination of the taboo search algorithm and the fuzzy C-means algorithm to cluster similar people in the sample to achieve automatic classification of the sample, effectively avoiding the shortcomings of the algorithm being easily affected by initial values ​​and outliers when the K-Means algorithm or the fuzzy C-means algorithm is used for clustering, and the clustering result may not be the global optimal but the local optimal, making the classification result more stable and reliable, and providing strong data support for the subsequent credit assessment model.

[0027] In this embodiment, step S2 specifically includes the following contents.

[0028] Step S21, obtaining a set of initial cluster centers by using the fuzzy C-means algorithm for data clustering operations in unsupervised learning , remember the current point , global optimal point , calculate the optimal fitness function value , leave the taboo table empty.

[0029] Specifically, the taboo table is a way to store visited solutions or operations in the taboo search algorithm to avoid repeated searches or falling into local optimal solutions. And a set of initial cluster centers are obtained through the fuzzy C-means algorithm, paving the way for subsequent clustering work.

[0030] In this embodiment, step S21 specifically includes the following contents.

[0031] Step S211, set the number of cluster centroids c and fuzzy index , initialize the centroids of each cluster , set the convergence accuracy , let the number of iterations n=0.

[0032] Specifically, the parameters in the algorithm are initialized first, where the number of cluster centroids c defines the number of clusters that are expected to be obtained in the end, which can be selected based on actual problems and experience. The fuzzy index m controls the fuzziness of the clustering, and is usually a real number greater than 1. The larger the value, the more fuzzy the membership. The centroids of each initialized cluster can be selected randomly.

[0033] Step S212, calculate samples To cluster centroid The Euclidean distance , update membership ;like < , then let ; Get a new set of membership .

[0034] Step S213, obtain a new set of cluster centroids based on the new membership If calculated , then the algorithm stops; otherwise, And jump to step S212 to continue execution.

[0035] Specifically, calculate each sample data point With each cluster center Euclidean distance , and update the membership of each data point to each cluster according to the calculated distance and the current fuzzy index m , usually, the distance The closer the membership The higher the membership The change is less than a certain threshold When , recalculate the center of each cluster to get a new set of cluster centroids Repeat the above steps until the convergence condition is reached , that is, the change is less than the set convergence accuracy, the algorithm is stopped to obtain the initial centroids of each cluster, otherwise the number of iterations n is increased to continue to repeat the above steps.

[0036] Step S22, determine whether the stop criterion is met, the stop criterion is configured to use the maximum number of iterations criterion, that is, iterate to a set number of times and stop, if it is met, terminate the algorithm process, otherwise the current point Randomly pick points in the neighborhood Second, based on the collected points, the fuzzy C-means algorithm is used to perform local optimization and generate Neighborhood solution sequence .

[0037] Specifically, after obtaining the initial cluster center, at the current point Randomly select several points in the neighborhood of as candidate points, execute the fuzzy mean algorithm for local optimization, and generate a better cluster center combination (i.e., neighborhood solution sequence) , this sequence contains all possible cluster center combinations found in the neighborhood of the current point, and stops the above steps when the maximum number of iterations is reached. The maximum number of iterations criterion is a common stopping criterion. When the algorithm has iterated a set number of times, regardless of whether the current solution is optimal, the algorithm will stop to prevent the algorithm from running indefinitely, especially when it is difficult to find the global optimal solution.

[0038] Step S23, the obtained neighborhood solution sequence , calculate the fitness function values ​​of all candidate solutions , and get the optimal solution And the corresponding adaptation value ,judge Whether the amnesty rules are met, if the amnesty rules are met, Higher than , update the global optimal point , the optimal fitness function value ; If the amnesty rules are not met, then judge Is it in a taboo state? If not, remember the current point. ; If it is in a taboo state, select the best solution from the solutions in the non-taboo state and record it as ; If all points are in the taboo state, take the linear average of all points in the taboo table as the new current solution .

[0039] Specifically, the amnesty rule is a special judgment condition used to decide whether to allow some solutions that do not meet the conventional selection conditions to be accepted as new global optimal points to avoid the algorithm falling into the local optimal solution. Then, calculate the fitness function values ​​of all candidate solutions To evaluate the pros and cons of each candidate solution and find the best solution And judge whether it satisfies the amnesty rule (that is, the adaptation function value Greater than the current adaptation function value ), if it satisfies, update the global optimal point and the optimal fitness function value, otherwise determine whether the solution is in a taboo state, if not in a taboo state, record the point as the current point For subsequent cycles, if it is in a taboo state, the optimal solution is selected from the solutions in the non-taboo state as the new current point. If all points are in a taboo state, the linear average of all points in the taboo table is taken as the new current point.

[0040] Step S24: Add it into the taboo table and update the taboo table according to the length of the taboo table, and return to step S22.

[0041] Step S25: After the stop criterion is met, the samples are classified based on the final cluster centers.

[0042] Specifically, the current point Add it to the taboo table and update it according to the length of the taboo table to ensure that the current point will not be selected again, avoid the algorithm falling into the local optimum, and return to step S22 for iterative optimization. Finally, after meeting the stopping criteria, classify the samples based on the final cluster center to obtain high-quality sample classification results.

[0043] Step S3, construct a dual cost-sensitive loss function, introduce a balance factor α to optimize the focus loss function, and adjust the optimization target of the tree model to the focus loss function for training.

[0044] Specifically, cost-sensitive learning is an important machine learning method. Its core idea is to take into account the different costs of different types of errors, and adjust the decision boundary of the classifier accordingly to minimize the total cost caused by incorrect classification. It adjusts the contribution of misjudgment of important category samples to the overall loss by adding a cost-sensitive factor to the loss function, so that the model pays more attention to the prediction of this type of samples, so as to improve the model's recognition ability for minority class samples or high-cost errors. However, since customer samples are divided into easy-to-classify customer samples and difficult-to-classify customer samples, there are often a large number of easy-to-classify customer samples and a small number of difficult-to-classify customer samples in the training set. Simply using traditional models combined with cost-sensitive factors cannot solve the problem that easy-to-classify customer samples dominate the model learning direction and difficult-to-classify customer samples in difficult areas. Therefore, this embodiment optimizes the focus loss function by introducing a balance factor α to construct a dual cost-sensitive loss function to increase the weights of misclassified samples and important samples. The optimized loss function is: ,in Focus on adjusting the weights of minority categories, Focus on adjusting the weights of misclassified samples. The weight selection is configured to be based on the Bayesian parameter adjustment strategy or the grid parameter adjustment strategy. Output the corresponding sample probability for the model, .

[0045] Step S4, using the preprocessed data, adding classification features, and adjusting the optimization objective function of the distributed gradient boosting decision tree XGBoost to a dual cost-sensitive loss function to train the personal credit assessment model.

[0046] Specifically, the XGBoost algorithm is an optimization algorithm based on the gradient boosting tree GBDT, which gradually reduces the prediction error of the model by training a new decision tree in each round of iteration, thereby constructing a strong classifier or regressor. In this embodiment, after the preprocessed sample data is classified and the label features are added, it is input into the XGBoost model with the constructed dual cost sensitive loss function as the optimization target, so that the model can output the corresponding personal credit assessment result according to the input personal information data.

[0047] Step S5, input the personal information data to be evaluated into the trained personal credit evaluation model to obtain the corresponding credit evaluation result.

[0048] Specifically, after the training is completed, the user's credit status can be evaluated based on the input personal information.

[0049] The personal credit assessment method based on the dual cost-sensitive loss function disclosed in this embodiment, after pre-processing and cleaning the training sample data, classifies similar sample populations by introducing a taboo search algorithm in the fuzzy C-means algorithm, thereby effectively improving the clustering quality and discrimination of the sample data. At the same time, a dual cost-sensitive loss function is constructed by introducing a balance factor in the focus loss function, and the optimization objective function of the distributed gradient boosting decision tree XGBoost is adjusted to the dual cost-sensitive loss function to consider the cost difference between different samples, thereby effectively dealing with the problem of sample imbalanced data sets and improving the model's ability to identify minority class samples or high-cost errors. Finally, the sample data is input into the personal credit assessment model for training, so that the model can obtain the corresponding credit assessment results according to the input personal information data, significantly improving the accuracy of the personal credit assessment model, and reducing the interference of the initial point setting, abnormal data, and sample imbalance on the model.

[0050] In another embodiment, as shown in the attached Figure 2As shown, the present invention also discloses a personal credit assessment system based on a dual cost-sensitive loss function, including a sample processing module 1, a sample classification module 2, a function optimization module 3, a model training module 4 and a result acquisition module 5. The sample processing module 1 is used to pre-process the data, including obtaining each personal information data and loan behavior time as a sample, matching the corresponding features, and labeling each sample through rolling rate analysis and account age analysis. The features include but are not limited to historical repayment behaviors and historical credit records of various institutions, and the personal information data includes ID cards, mobile phone numbers and names. The sample classification module 2 is used to combine the taboo search algorithm and the fuzzy C-means algorithm to construct category features to classify similar population samples. The function optimization module 3 is used to construct a dual cost-sensitive loss function, introduce a balance factor α to optimize the focus loss function, and adjust the optimization target of the tree model to the focus loss function for training. The model training module 4 is used to use the pre-processed data, add classification features, and adjust the optimization objective function of the distributed gradient boosting decision tree XGBoost to the dual cost-sensitive loss function to train the personal credit assessment model. The result acquisition module 5 is used to input the personal information data to be evaluated into the trained personal credit evaluation model to obtain the corresponding credit evaluation result.

[0051] In this embodiment, the sample classification module specifically includes an initial configuration module, a local optimization module, an optimal solution module, a taboo update module and a sample classification module, wherein the initial configuration module is used to obtain a set of initial cluster centers by using a fuzzy C-means algorithm for data clustering operations in unsupervised learning. , remember the current point , global optimal point , calculate the optimal fitness function value , set the taboo table to empty; local optimization module, used to determine whether the stopping criterion is met, the stopping criterion is configured to use the maximum number of iterations criterion, that is, iterate to the set number of times to stop, if it is satisfied, the algorithm process is terminated, otherwise the current point Randomly pick points in the neighborhood Second, based on the collected points, the fuzzy C-means algorithm is used to perform local optimization and generate Neighborhood solution sequence ; Optimal solution module, used to obtain the neighborhood solution sequence , calculate the fitness function values ​​of all candidate solutions , and get the optimal solution And the corresponding adaptation value ,judge Whether the amnesty rules are met, if the amnesty rules are met, Higher than , update the global optimal point , the optimal fitness function value ; If the amnesty rules are not met, then judge Is it in a taboo state? If not, remember the current point. ; If it is in a taboo state, select the best solution from the solutions in the non-taboo state and record it as ; If all points are in the taboo state, take the linear average of all points in the taboo table as the new current solution ; Taboo update module, used to Add to the taboo table and update the taboo table according to the length of the taboo table, and return to the local optimization module; the sample classification module is used to classify the samples based on the final cluster center after meeting the stopping criteria.

[0052] In this embodiment, the initial configuration module further includes a parameter configuration module, a membership updating module and a cluster centroid updating module, wherein the parameter configuration module is used to set the number of cluster centroids c and the fuzzy index , initialize the centroids of each cluster , set the convergence accuracy , record the number of iterations n=0; membership update module, used to calculate the sample To cluster centroid The Euclidean distance , update the membership ;like < , then let ; Get a new set of membership ; Cluster centroid update module, used to obtain a new set of cluster centroids based on the new membership If calculated , then the algorithm stops; otherwise, And jump to step S212 to continue execution.

[0053] In this embodiment, the loss function is: ,in Focus on adjusting the weights of minority categories, Focus on adjusting the weights of misclassified samples. The weight selection is configured to be based on the Bayesian parameter adjustment strategy or the grid parameter adjustment strategy.

[0054] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and similar parts between the various embodiments can be referred to each other. For the personal credit assessment system based on the dual cost-sensitive loss function disclosed in the embodiment, since it corresponds to the personal credit assessment method based on the dual cost-sensitive loss function disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the aforementioned method part.

[0055] In other embodiments, a server is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the personal credit assessment method based on the dual cost-sensitive loss function as described in the above embodiments when executing the computer program. The server may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a server and does not constitute a limitation on the server, and may include more or fewer components than shown in the diagram, or combine certain components, or different components.

[0056] If the server is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned individual credit assessment method embodiments based on the dual cost-sensitive loss function can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0057] In conclusion, the above is only a preferred embodiment of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the patent of the present invention.

Claims

1. A personal credit assessment method based on a dual cost-sensitive loss function, characterized in that: The steps include: S1, preprocessing the data, including obtaining the personal information data and loan behavior time of each sample, matching the corresponding features, and labeling each sample through rolling rate analysis and aging analysis. The features include but are not limited to historical repayment behavior and historical credit records of each institution. The personal information data includes ID card, mobile phone number and name; S2, combining the taboo search algorithm and the fuzzy C-means algorithm to construct category features to classify similar population samples; S3, construct a dual cost-sensitive loss function, introduce a balance factor α to optimize the focus loss function, and adjust the optimization target of the tree model to the focus loss function for training; S4, using the preprocessed data, adding classification features, and adjusting the optimization objective function of the distributed gradient boosting decision tree XGBoost to a dual cost-sensitive loss function to train the personal credit assessment model; S5, inputting the personal information data to be evaluated into the trained personal credit evaluation model to obtain the corresponding credit evaluation result.

2. The personal credit assessment method based on dual cost-sensitive loss function according to claim 1 is characterized in that: The step S2 specifically includes: S21, obtain a set of initial cluster centers through the fuzzy C-means algorithm for data clustering operations in unsupervised learning , remember the current point , global optimal point , calculate the optimal fitness function value , leave the taboo table empty; S22, determine whether the stopping criterion is met, the stopping criterion is configured to use the maximum number of iterations criterion, that is, iterate until the set number of times stops, if it is met, terminate the algorithm process, otherwise, Randomly pick points in the neighborhood Second, based on the collected points, the fuzzy C-means algorithm is used to perform local optimization and generate Neighborhood solution sequence ; S23, the obtained neighborhood solution sequence , calculate the fitness function values ​​of all candidate solutions , and get the optimal solution And the corresponding adaptation value ,judge Whether the amnesty rules are met, if the amnesty rules are met, Higher than , update the global optimal point , the optimal fitness function value ; If the amnesty rules are not met, then judge Is it in a taboo state? If not, remember the current point. ; If it is in a taboo state, select the best solution from the solutions in the non-taboo state and record it as ; If all points are in the taboo state, take the linear average of all points in the taboo table as the new current solution ; S24, will Add the item to the taboo list and update the taboo list according to the length of the taboo list, and return to step S22; S25, after the stopping criterion is met, the samples are classified based on the final cluster centers.

3. The personal credit assessment method based on dual cost-sensitive loss function according to claim 2 is characterized in that: The step S21 specifically includes: S211, set the number of cluster centroids c and fuzzy index , initialize the centroids of each cluster , set the convergence accuracy , let the number of iterations n=0; S212, calculation sample To cluster centroid The Euclidean distance , update the membership ;like < , then let ; Get a new set of membership ; S213, obtain a new set of cluster centroids based on the new membership If calculated , then the algorithm stops; otherwise, And jump to step S212 to continue execution.

4. The personal credit assessment method based on dual cost-sensitive loss function according to claim 1 is characterized in that: The loss function is: ,in Focus on adjusting the weights of minority categories, Focus on adjusting the weights of misclassified samples. The weight selection is configured to be based on the Bayesian parameter adjustment strategy or the grid parameter adjustment strategy. Output the corresponding sample probability for the model, .

5. A personal credit assessment system based on a dual cost-sensitive loss function, characterized in that: include: The sample processing module is used to pre-process the data, including obtaining the personal information data and loan behavior time of each sample, matching the corresponding features, and labeling each sample through rolling rate analysis and aging analysis. The features include but are not limited to historical repayment behavior and historical credit records of each institution. The personal information data includes ID card, mobile phone number and name; The sample classification module is used to combine the taboo search algorithm and the fuzzy C-means algorithm to construct category features to classify samples of similar people; Function optimization module, which is used to construct a dual cost-sensitive loss function, introduce a balance factor α to optimize the focus loss function, and adjust the optimization target of the tree model to the focus loss function for training; The model training module is used to use the preprocessed data, add classification features, and adjust the optimization objective function of the distributed gradient boosting decision tree XGBoost to a dual cost-sensitive loss function to train the personal credit assessment model; The result acquisition module is used to input the personal information data to be evaluated into the trained personal credit evaluation model and obtain the corresponding credit evaluation results.

6. The personal credit evaluation system based on dual cost-sensitive loss function according to claim 5 is characterized in that: The sample classification module specifically includes: Initial configuration module, used to obtain a set of initial cluster centers through the fuzzy C-means algorithm for data clustering operations in unsupervised learning , remember the current point , global optimal point , calculate the optimal fitness function value , leave the taboo table empty; The local optimization module is used to determine whether the stopping criterion is met. The stopping criterion is configured to use the maximum number of iterations criterion, that is, the algorithm process is terminated if it is met, otherwise the current point is Randomly pick points in the neighborhood Second, based on the collected points, the fuzzy C-means algorithm is used to perform local optimization and generate Neighborhood solution sequence ; The optimal solution module is used to obtain the neighborhood solution sequence , calculate the fitness function values ​​of all candidate solutions , and get the optimal solution And the corresponding adaptation value ,judge Whether the amnesty rules are met, if the amnesty rules are met, Higher than , update the global optimal point , the optimal fitness function value ; If the amnesty rules are not met, then judge Is it in a taboo state? If not, remember the current point. ; If it is in a taboo state, select the best solution from the solutions in the non-taboo state and record it as ; If all points are in the taboo state, take the linear average of all points in the taboo table as the new current solution ; Taboo update module, used to Add the item to the taboo list and update the taboo list according to the length of the taboo list, and return to step S22; The sample classification module is used to classify samples based on the final cluster centers after the stopping criteria are met.

7. The personal credit evaluation system based on dual cost-sensitive loss function according to claim 6 is characterized in that: The initial configuration module also includes: Parameter configuration module, used to set the number of cluster centroids c and fuzzy index , initialize the centroids of each cluster , set the convergence accuracy , let the number of iterations n=0; Membership update module, used to calculate samples To cluster centroid The Euclidean distance , update the membership ;like < , then let ; Get a new set of membership ; Cluster centroid update module, used to obtain a new set of cluster centroids based on the new membership If calculated , then the algorithm stops; otherwise, And jump to step S212 to continue execution.

8. The personal credit evaluation system based on dual cost-sensitive loss function according to claim 7 is characterized in that: The loss function is: ,in Focus on adjusting the weights of minority categories, Focus on adjusting the weights of misclassified samples. The weight selection is configured to be based on the Bayesian parameter adjustment strategy or the grid parameter adjustment strategy. Output the corresponding sample probability for the model, .

9. A server comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Citation Information

Patent Citations

  • Mobile application behavior dynamic credibility assessment method and device

    CN109657452A

  • Internet credit assessment method for individual households and small and micro enterprise e-commerce merchants

    CN117273919A