Credit line determination method and system, electronic equipment and storage medium
The default risk assessment model and quota determination model are constructed through the GBDT algorithm, which solves the problem that traditional credit line determination methods ignore dynamic changes, and achieves more accurate and scientific credit line determination, reducing risks and improving customer experience.
Patent Information
- Application Number
- CN202411907684.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-04-18
AI Technical Summary
Traditional credit line determination methods ignore the dynamic changes in customer income volatility, consumption habits and credit behavior, resulting in the credit line that may be too high or too low, affecting the risk management and customer experience of financial institutions.
Multiple decision trees are built using the GBDT algorithm, and a default risk assessment model is constructed based on the customer's personal social credit sample data, and a quota determination model is constructed based on the income level and total assets, and the credit line is accurately determined.
By comprehensively considering customers' debt repayment ability and credit risks, avoiding dependence on a single indicator, making the credit line determination more in line with the actual situation of the customer, reasonably allocate credit resources, reduce risks, and improve the scientificity and reliability of decision-making.
Smart Images
Figure CN119850325A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of financial credit technology, and more specifically to a credit limit determination method and system, electronic equipment and storage medium. Background Art
[0002] At present, in financial credit business, it is crucial to reasonably and accurately determine the credit limit for customers. At present, the traditional way of determining credit limits often has many limitations. For example, some financial institutions simply set the credit limit range based on the customer's static financial data (such as income, assets, etc.), ignoring factors such as the volatility of customer income, consumption habits, and dynamic changes in credit behavior. As a result, the determined credit limit may be too high, exposing financial institutions to a greater risk of default; or too low, unable to meet the customer's reasonable credit needs, affecting customer experience and business expansion. In addition, some methods lack a scientific and effective weight allocation mechanism when comprehensively considering multiple factors, so that the impact of each factor on the determination of the credit limit cannot be accurately reflected.
[0003] Therefore, how to build a more intelligent, accurate and comprehensive credit limit determination plan to improve the scientificity and rationality of credit business is an issue that technical personnel in this field urgently need to solve. Summary of the invention
[0004] In view of this, the present invention provides a credit limit determination method and system, an electronic device and a storage medium to solve the problems existing in the background technology.
[0005] In order to achieve the above object, the present invention adopts the following technical solution:
[0006] A method for determining a credit limit, comprising:
[0007] Collect personal social credit sample data of customers;
[0008] Preprocess the collected personal social credit sample data;
[0009] The GBDT algorithm is used to construct multiple decision trees, each of which is fitted on the residual of the previous tree to build a default risk assessment model; the preprocessed personal social credit sample data is input into the default risk assessment model to output the default risk probability value;
[0010] Construct a credit limit determination model based on the default risk probability value output by the default risk assessment model and the customer's income level and total assets;
[0011] The credit limit determination model is used to determine the credit limit of the customer to be determined.
[0012] Optionally, the personal social credit sample data includes the customer's identity information, financial status information, historical credit records, consumption behavior data, and social network data.
[0013] Optionally, the preprocessing includes deleting missing data and erroneous data, and preprocessing the data using a normalization method to obtain preprocessed data.
[0014] Optionally, the specific steps of constructing the default risk assessment model are:
[0015] (1) Determine input: Determine the training set based on the preprocessed personal social credit sample data;
[0016] (2) Determine the output: default risk assessment model, output the default risk probability value;
[0017] (3) Initialization: constructing the initial default risk assessment model;
[0018] (4) For each sample data, calculate the residual; fit a new regression tree based on the calculated residual;
[0019] (5) Update the current learner to obtain the discriminant function when the input data belongs to the leaf node area; and obtain the final default risk assessment model based on the discriminant function.
[0020] Optionally, training a default risk assessment model includes:
[0021] Training of a default risk assessment model based on a gradient boosting decision tree; wherein, a 10-fold cross-validation and grid search method is used for parameter optimization. First, the range of parameter combinations is determined by grid search, and then the model is evaluated in combination with 10-fold cross-validation to find the optimal parameter combination. The parameters include: the number of decision trees, the learning rate of the model, the sub-sample ratio, and the depth limit of each decision tree.
[0022] Optionally, perform customer sample screening and generate a quota determination model data set, including:
[0023] Screening the customer samples according to customer sample screening conditions to obtain target customer samples, wherein the customer sample screening conditions include: the customer's default risk probability value, the customer's income level, and the customer's total assets;
[0024] The training set and test set in the target customer sample are determined according to the preset ratio.
[0025] Optionally, the step of constructing a quota determination model specifically includes:
[0026] The linear regression model established by the ridge regression method is trained according to the modeling variables in the training set and the test set in the target customer sample to obtain the test result value, significance level and regression coefficient of the modeling variable in the linear regression model; when the test result value and significance level meet the conditions, the credit limit determination model is obtained through the trained linear regression model;
[0027] A regularization term is added to the loss function of the linear regression model, and the linear regression model is:
[0028] y(a,w)=w1a1+w2a2+…+w n a n
[0029] Among them, a is the modeling variable, and w is the regression coefficient of the linear regression model;
[0030] The loss function with the regularization term added is:
[0031]
[0032] Where n is the number of modeling variables in the training set, is a regular term, λ is the sum of squares of the regression coefficients of each modeling variable, and a i is the i-th modeling variable in the training set, y i is the credit limit data information of the i-th sample in the training set, and w is the regression coefficient of the linear regression model.
[0033] A credit limit determination system, comprising:
[0034] Data collection module, collecting customers’ personal social credit sample data;
[0035] A preprocessing module is used to preprocess the collected personal social credit sample data;
[0036] The risk assessment model building module uses the GBDT algorithm to build multiple decision trees. Each tree is fitted on the residual of the previous tree to build a default risk assessment model. The pre-processed personal social credit sample data is input into the default risk assessment model to output the default risk probability value.
[0037] A credit limit determination model building module is used to build a credit limit determination model based on the default risk probability value output by the default risk assessment model;
[0038] The result output module uses the credit limit determination model to determine the credit limit of the customer to be determined.
[0039] An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein:
[0040] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the above method.
[0041] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.
[0042] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a method and system for determining a credit limit, an electronic device and a storage medium, firstly collects the personal social credit sample data of the customer, then pre-processes the collected data, and then uses the GBDT algorithm to construct multiple decision trees, each tree is fitted on the residual of the previous tree to construct a default risk assessment model, and the pre-processed personal social credit sample data is input into the model to obtain the default risk probability value, and a limit determination model is constructed based on the default risk probability value and the customer's income level and total assets, and finally the credit limit of the customer to be determined is determined using the limit determination model. The present invention comprehensively considers the customer's debt repayment ability and credit risk, avoids the limitation of relying on only a single indicator, makes the credit limit determination more in line with the customer's actual situation, reasonably allocates credit resources, reduces credit risks, improves the scientificity and reliability of credit business decisions, and ensures the financial security and operational efficiency of financial institutions. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0044] Figure 1 A flow chart of the method provided by the present invention.
[0045] Figure 2 This is a system structure diagram provided by the present invention. DETAILED DESCRIPTION
[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0047] The embodiment of the present invention discloses a method for determining a credit limit. Figure 1 As shown, including:
[0048] Collect personal social credit sample data of customers;
[0049] Preprocess the collected personal social credit sample data;
[0050] The GBDT algorithm is used to construct multiple decision trees, each of which is fitted on the residual of the previous tree to build a default risk assessment model; the preprocessed personal social credit sample data is input into the default risk assessment model to output the default risk probability value;
[0051] Construct a credit limit determination model based on the default risk probability value output by the default risk assessment model and the customer's income level and total assets;
[0052] The credit limit determination model is used to determine the credit limit of the customer to be determined.
[0053] In a specific embodiment, the personal social credit sample data includes the customer's identity information, financial status information, historical credit records, consumption behavior data, and social network data.
[0054] Specifically, data from many aspects is collected, including but not limited to the customer's basic identity information (age, occupation, etc.), financial status information (income, assets, liabilities, etc.), historical credit records (past loan repayments, credit card usage and repayment records, etc.), consumer behavior data (daily consumer categories, consumption frequency, consumption amount, etc.) and social network data (social activity, correlation between social circle credit level, etc.). The above data is obtained by connecting with multiple data sources, such as the bank's internal customer database, third-party payment platform, credit reporting agency, etc. Taking a commercial bank as an example, first, the bank obtains the customer's basic identity information and financial status information through the bank's internal customer information management system, and at the same time cooperates with the third-party payment platform to obtain the customer's consumer behavior data for the past year, and then obtains the customer's complete historical credit record from the credit reporting agency.
[0055] In a specific embodiment, the preprocessing includes deleting missing data and erroneous data, and preprocessing the data using a normalization method to obtain preprocessed data. The method here can be processed using an existing method.
[0056] In a specific embodiment, the specific steps of constructing the default risk assessment model are:
[0057] (1) Determine input: Determine the training set based on the preprocessed personal social credit sample data;
[0058] Training set Where X is the input data set, Y is the default risk score set, xi is the i-th element in the input data set (i.e., the i-th input data), i ranges from 1 to N, N is the total amount of data, R n represents n-dimensional real number space, R is one-dimensional real number space. The i-th input data x i V is a vector composed of the customer's identity information, financial information, historical credit record, consumer behavior data, and social network data. x ,y i For x i The default risk score corresponding to y is L(y,f(x)), where y is the collected default risk score and f(x) is the default risk score output by the model prediction corresponding to y.
[0059] (2) Determine the output: default risk assessment model Output the default risk probability value;
[0060] (3) Initialization: constructing the initial default risk assessment model;
[0061]
[0062] Among them, f0(x) is the initial default risk assessment model, Indicates the value of c that reaches the minimum value, c is y i The corresponding predicted value, L(y i ,c) is the loss function.
[0063] (4) For each sample data, calculate the residual; fit a new regression tree based on the calculated residual; for m = 1, 2, ..., M, where M is the number of iterations;
[0064] For each sample (i.e. each input data) x i (i=1,2,…,N), calculate the residual r mi (The residual here is the loss function L(y i ,f(x i )) is approximated by the negative gradient value:
[0065]
[0066] Based on the residual r of the previous step mi Fit a new regression tree (i.e., decision tree), whose leaf node area is denoted by R mj ,j=1,2,…,J, where J represents the number of leaf nodes of the mth decision tree. The number of decision trees is equal to the number of iterations. L(y i ,f(x i )) is the loss function, f(x i ) is the i-th input data x iThe corresponding model predicts the default risk score output, f(x) = f m-1 (x) is the value of f m-1 (x).
[0067] For j = 1, 2, ..., J, calculate:
[0068]
[0069] In the formula, c mj for The value of c when it reaches the minimum value, f m-1 (x i ) is the default risk score output by the model prediction in the previous step.
[0070] (5) Update the current learner to obtain the discriminant function when the input data belongs to the leaf node area; and obtain the final default risk assessment model based on the discriminant function.
[0071] Update the current learner and get:
[0072]
[0073] In the formula, I(x∈R mj ) is the input data x belonging to the leaf node region R mj The discriminant function of .
[0074] The final coal mine risk score prediction model is obtained:
[0075]
[0076] In a specific embodiment, training the default risk assessment model includes:
[0077] The main parameters involved in the training process of the GBDT model are as follows:
[0078] (1)n_estimators
[0079] n_estimators represents the number of decision trees, that is, the number of model iterations. Generally speaking, the larger the n_estimators value, the stronger the model fitting ability, but too large a n_estimators value may cause the model to overfit.
[0080] Therefore, when selecting n_estimators, it is necessary to consider the model's fit and efficiency. The default value of n_estimators is 100.
[0081] (2) learning_rate
[0082] learning_rate represents the learning rate of the model. The learning rate represents the contribution weight of each base learner in the entire iteration. The default value is generally 0.1, which is used to control the step size of each tree model update, also known as the step size factor. Generally speaking, the smaller the learning rate, the slower the model converges, and more iterations are required to achieve the desired effect. Therefore, when adjusting the learning rate, it is necessary to comprehensively consider the number of iterations and the size of the learning rate to determine the appropriate parameter combination.
[0083] (3)subsample
[0084] Subsample is used to control the subsample ratio used by each base learner, and to prevent overfitting by sampling samples. The value range of subsample is (0,1], and it is generally recommended to be between [0.5,0.8]. The default value is 1. When the value is 1, it means that during the training process of the model, the training set used each time is the same as the original sample set, that is, no sample is sampled.
[0085] (4)max_depth
[0086] max_depth indicates the depth limit of each decision tree, which is used to control the complexity of the model. When max_depth is small, the model's fitting ability is weak and it is easy to underfit; when max_depth is too large, the model is easy to overfit. The default value of max_depth is 3. When the depth is 1, the generated trees are all binary trees with only two leaf nodes, so there is no interaction between the variables. When the depth is 2, the generated tree model has interaction between two variables. By analogy, deeper base trees can explore very complex variable interactions, but they are also more likely to cause overfitting.
[0087] In the process of model parameter adjustment, the K-fold cross-validation method is usually used to find the optimal parameters. K-fold cross-validation is a commonly used model evaluation technique, which divides the data set into K non-overlapping equal subsets, and uses K training and testing to evaluate the performance of the model. In each cross-validation, one subset is used as a test set, and the remaining K-1 subsets are used as training sets to train the model, and the final result is the average value of all cross-validation results. The present invention uses a 10-fold cross-validation and grid search method to perform parameter optimization. First, the range of parameter combinations is determined by grid search, and then the model is evaluated in combination with 10-fold cross-validation to find the optimal parameter combination.
[0088] In a specific embodiment, customer sample screening is performed to generate a credit limit determination model data set, including:
[0089] Screen the customer samples according to the customer sample screening conditions to obtain the target customer samples. The customer sample screening conditions include: the customer's default risk probability value, the customer's income level, and the customer's total assets;
[0090] The training set and test set in the target customer sample are determined according to the preset ratio.
[0091] Specifically, according to a preset ratio, for example, 7:3, 70% of the collection is used as a training set, and the remaining 30% of the collection is used as a test set. It should be noted that those skilled in the art can choose to set the ratio according to actual conditions, and the ratio is not limited in this embodiment, and the ratio can be 9:1, 8:2, 6:4, 5:5, etc.
[0092] In a specific embodiment, constructing a quota determination model specifically includes:
[0093] The linear regression model established by the ridge regression method is trained according to the modeling variables in the training set and the test set in the target customer sample to obtain the test result value, significance level and regression coefficient of the modeling variable in the linear regression model;
[0094] The test result value can usually be a t value, which is usually an indicator of significance. The significance level usually refers to the p value. The t value and the p value are indicators for measuring whether the modeling variable is effective. The modeling variables are included in both the test set of the target object and the training set of the target object.
[0095] When the test result value and significance level meet the conditions, the quota determination model is obtained through the trained linear regression model;
[0096] Add a regular term to the loss function of the linear regression model, and the linear regression model is:
[0097] y(a,w)=w1a1+w2a2+…+w n a n
[0098] Among them, a is the modeling variable, and w is the regression coefficient of the linear regression model;
[0099] The loss function with the regularization term added is:
[0100]
[0101] Where n is the number of modeling variables in the training set, is a regular term, λ is the sum of squares of the regression coefficients of each modeling variable, and a i is the i-th modeling variable in the training set, y iis the credit limit data information of the i-th sample in the training set, and w is the regression coefficient of the linear regression model.
[0102] In this embodiment, by introducing the L2 regularization term, when solving the regression coefficient, it is necessary to make the loss function as small as possible, so that the coefficients of each modeling variable become averaged, and prevent the loss function from becoming larger due to the regression coefficient of a certain modeling variable being too large. As a result, the obtained credit limit determination model will not cause inaccurate credit limit model calculation due to the regression coefficient of a certain modeling variable being too large, and it can be ensured that the credit limit model comprehensively considers the information of each modeling variable.
[0103] A system for determining credit limits, such as Figure 2 As shown, including:
[0104] Data collection module, collecting customers’ personal social credit sample data;
[0105] A preprocessing module is used to preprocess the collected personal social credit sample data;
[0106] The risk assessment model building module uses the GBDT algorithm to build multiple decision trees. Each tree is fitted on the residual of the previous tree to build a default risk assessment model. The pre-processed personal social credit sample data is input into the default risk assessment model to output the default risk probability value.
[0107] A credit limit determination model building module is used to build a credit limit determination model based on the default risk probability value output by the default risk assessment model;
[0108] The result output module uses the credit limit determination model to determine the credit limit of the customer to be determined.
[0109] An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein:
[0110] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor so that the at least one processor can perform the steps of the above methods.
[0111] A computer storage medium stores a computer program, which implements the steps of the above methods when executed by a processor.
[0112] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in the present disclosure may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.
[0113] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0114] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for determining a credit limit, characterized in that: include: Collect personal social credit sample data of customers; Preprocess the collected personal social credit sample data; The GBDT algorithm is used to construct multiple decision trees, each of which is fitted on the residual of the previous tree to build a default risk assessment model; the preprocessed personal social credit sample data is input into the default risk assessment model to output the default risk probability value; Construct a credit limit determination model based on the default risk probability value output by the default risk assessment model and the customer's income level and total assets; The credit limit determination model is used to determine the credit limit of the customer to be determined.
2. A method for determining a credit limit according to claim 1, characterized in that: The personal social credit sample data includes the customer's identity information, financial status information, historical credit records, consumption behavior data and social network data.
3. A method for determining a credit limit according to claim 1, characterized in that: The preprocessing includes deleting missing data and erroneous data, and preprocessing the data using a normalization method to obtain preprocessed data.
4. A method for determining a credit limit according to claim 1, characterized in that: The specific steps of constructing the default risk assessment model are: (1) Determine input: Determine the training set based on the preprocessed personal social credit sample data; (2) Determine the output: default risk assessment model, output the default risk probability value; (3) Initialization: constructing the initial default risk assessment model; (4) For each sample data, calculate the residual; fit a new regression tree based on the calculated residual; (5) Update the current learner to obtain the discriminant function when the input data belongs to the leaf node area; The final default risk assessment model is obtained based on the discriminant function.
5. A method for determining a credit limit according to claim 4, characterized in that: Training the default risk assessment model includes: Training of a default risk assessment model based on a gradient boosting decision tree; wherein, a 10-fold cross-validation and grid search method is used for parameter optimization. First, the range of parameter combinations is determined by grid search, and then the model is evaluated in combination with 10-fold cross-validation to find the optimal parameter combination. The parameters include: the number of decision trees, the learning rate of the model, the sub-sample ratio, and the depth limit of each decision tree.
6. A method for determining a credit limit according to claim 1, characterized in that: Screen customer samples and generate a quota determination model data set, including: Screening the customer samples according to customer sample screening conditions to obtain target customer samples, wherein the customer sample screening conditions include: the customer's default risk probability value, the customer's income level, and the customer's total assets; The training set and test set in the target customer sample are determined according to the preset ratio.
7. A method for determining a credit limit according to claim 6, characterized in that: The construction of the quota determination model specifically includes: The linear regression model established by the ridge regression method is trained according to the modeling variables in the training set and the test set in the target customer sample to obtain the test result value, significance level and regression coefficient of the modeling variable in the linear regression model; when the test result value and significance level meet the conditions, the credit limit determination model is obtained through the trained linear regression model; A regularization term is added to the loss function of the linear regression model, and the linear regression model is: y(a,w)=w1a1+w2a2+…+w n a n Among them, a is the modeling variable, and w is the regression coefficient of the linear regression model; The loss function with the regularization term added is: Where n is the number of modeling variables in the training set, is a regular term, λ is the sum of squares of the regression coefficients of each modeling variable, and a i is the i-th modeling variable in the training set, y i is the credit limit data information of the i-th sample in the training set, and w is the regression coefficient of the linear regression model.
8. A credit limit determination system, characterized in that: A method for determining a credit limit according to any one of claims 1 to 7 is applied, comprising: Data collection module, collecting customers’ personal social credit sample data; A preprocessing module is used to preprocess the collected personal social credit sample data; The risk assessment model building module uses the GBDT algorithm to build multiple decision trees. Each tree is fitted on the residual of the previous tree to build a default risk assessment model. The pre-processed personal social credit sample data is input into the default risk assessment model to output the default risk probability value. A credit limit determination model building module is used to build a credit limit determination model based on the default risk probability value output by the default risk assessment model; The result output module uses the credit limit determination model to determine the credit limit of the customer to be determined.
9. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform a credit limit determination method as claimed in any one of claims 1 to 7.
10. A computer storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a credit limit determination method according to any one of claims 1 to 7 is implemented.