Credit limit adjustment method based on RFM and decision tree model

Abstract

The invention relates to the technical field of credit evaluation, in particular to an RFM and decision tree model-based credit limit adjustment method, which comprises the following steps of: processing and sorting customer expression variable data corresponding to each dimension, and performing slice classification according to each preset slice category; each category variable corresponding to each dimension is formed; taking a plurality of category variables with large information increments corresponding to the corresponding dimensions in each slice category as dimension evaluation variables of the corresponding dimensions; calculating a customer value score corresponding to each customer according to the dimension evaluation variable corresponding to each dimension; determining a customer value grade corresponding to each customer; according to the determined customer value level corresponding to each customer, determining a customer final quota level corresponding to each customer; and according to the customer final quota level corresponding to each customer, determining the corresponding credit quota adjustment proportion of each customer under the corresponding customer final quota level.

Description

Technical Field

[0001] The present invention relates to the technical field of credit assessment, and in particular to a credit limit adjustment method based on RFM and decision tree models. Background Art

[0002] In the financial credit sector, properly adjusting a customer's credit limit is crucial. On the one hand, excessively high credit limits can expose financial institutions to default risk, leading to an increase in nonperforming loans and threatening their financial security. On the other hand, excessively low credit limits can restrict customers' normal funding needs, negatively impacting their experience and even leading to customer churn, hindering the financial institution's business expansion.

[0003] Traditional credit limit adjustment methods have significant shortcomings. Most rely heavily on manual experience and are highly subjective, making it difficult to comprehensively and objectively assess a customer's true credit status and value. Furthermore, these traditional methods employ relatively limited assessment metrics, often focusing solely on the customer's underlying risk, while ignoring the rich data surrounding the customer's borrowing behavior, such as the timing, amount, and frequency of borrowing. This results in inaccurate customer evaluations and an inability to adjust credit limits in a timely and flexible manner based on changes in customer behavior. In other words, existing credit adjustment methods rely on a single assessment metric, failing to ensure reliable and accurate adjustments to customer credit limits. Summary of the Invention

[0004] One of the purposes of the present invention is to provide a credit limit adjustment method based on RFM and decision tree model to solve the problem in the prior art that the reliability and accuracy of customer credit limit adjustment are low due to the single evaluation dimension, thereby achieving efficient and reliable adjustment of customer credit limit.

[0005] In order to achieve the above purpose, a credit limit adjustment method based on RFM and decision tree model is provided, which includes the following steps:

[0006] S1. Obtaining customer performance variable data corresponding to the R, F, and M dimensions, processing and organizing the customer performance variable data corresponding to each dimension, and slicing and classifying them according to preset slicing categories to form category variables corresponding to each dimension; the R, F, and M dimensions are the most recent loan time dimension, the loan frequency dimension, and the loan amount dimension, respectively;

[0007] S2. Based on the preset variable selection strategy and the categorical variables corresponding to each dimension, several categorical variables with large information increments corresponding to the corresponding dimensions in each slice category are used as dimension evaluation variables for the corresponding dimensions;

[0008] S3. Calculate the customer value score corresponding to each customer based on the dimensional evaluation variables corresponding to each dimension and the preset customer value score calculation strategy;

[0009] S4. Determine the customer value level corresponding to each customer based on the customer value score corresponding to each customer and a preset customer value score corresponding level table;

[0010] S5. Determine the final credit limit level for each customer based on the determined customer value level for each customer and other customer indicator information for each customer;

[0011] S6. According to the final credit limit level corresponding to each customer and based on a preset credit limit adjustment table, determine the credit limit adjustment ratio corresponding to each customer under the corresponding final credit limit level.

[0012] The technical principles and effects of this solution: This solution collects performance variable data on dimensions such as the time of the customer's most recent loan, the frequency of the loan, and the amount of the loan. This raw data is then processed and organized through data cleaning (handling missing values and outliers) and data transformation (such as standardization and normalization). The data is then sliced and classified according to pre-set slicing categories (such as time intervals and amount intervals), converting the data into different categorical variables within each dimension, providing a structured data foundation for subsequent analysis.

[0013] Based on the principles of decision tree models, we use pre-defined variable selection strategies (such as calculating information gain). For each dimension, we calculate the information increment (e.g., information gain) of each categorical variable to measure its importance to the target variable (e.g., customer value rating). The greater the information increment, the greater the variable's ability to distinguish customer value. From each slice category, we select several categorical variables with large information increments as dimension evaluation variables for the corresponding dimension. These variables will be used in subsequent customer value assessments.

[0014] Based on the selected dimension evaluation variables and the preset customer value score calculation strategy, a comprehensive customer value score is calculated for each customer. This score reflects the customer's comprehensive value level under the RFM model.

[0015] The calculated customer value score is matched against a preset customer value score grading table to determine each customer's customer value grading. For example, the scoring range can be divided into different intervals, with each interval corresponding to a value grading. Based on the known customer value grading, combined with other customer indicators (such as the customer's credit risk level, debt situation, etc.), a comprehensive assessment is conducted to determine each customer's final credit limit tier. This other indicator information can provide more comprehensive customer risk and value information, helping banks to more accurately judge a customer's credit limit tolerance. Finally, based on the customer's final credit limit tier, the preset credit limit adjustment table is referenced to determine the corresponding credit limit adjustment ratio for each customer at that credit limit tier. The credit limit adjustment table can be set according to the bank's credit policy and risk appetite, with different credit limit tiers corresponding to different adjustment ratios, thereby achieving reasonable adjustments to customer credit limits.

[0016] In today's highly competitive financial market, accurate customer value assessment and effective management are key to enhancing banks' competitiveness. This solution innovatively combines the RFM model with the decision tree model, demonstrating superior technical results in customer value assessment and management.

[0017] Traditional single-dimensional customer value assessment methods, such as those based solely on customer risk, fail to fully reflect a customer's true value. The RFM model comprehensively considers customer behavior across three dimensions: the time of last consumption (R), consumption frequency (F), and consumption amount (M). For example, in a credit scenario, the time of a customer's most recent loan reflects the timeliness of their funding needs, the frequency of borrowing reflects their reliance on credit services, and the amount of the loan indicates the scale of funding required. By cross-analyzing these three dimensions, banks can create a three-dimensional portrait of customer behavior, avoiding the one-sidedness that can result from single-dimensional assessments.

[0018] The RFM model quantitatively analyzes customer behavior across three dimensions: most recent purchase, frequency, and amount. For example, in the credit context, the R dimension reflects the timeliness of a customer's borrowing, the F dimension reflects borrowing activity, and the M dimension indicates the size of the loan. This data provides foundational information for customer value assessment and outlines the general outline of customer behavior.

[0019] Decision tree models uncover deep relationships: By learning from massive amounts of customer data, decision tree models uncover the complex relationships between various RFM dimensions and customer value. For example, the model can discover that customers whose borrowing frequency and amount meet a certain ratio within a specific time period tend to have higher value potential. By accurately understanding these relationships, the model can provide a more in-depth and accurate assessment of customer value.

[0020] This solution, through the combination of RFM and decision tree models, comprehensively assesses customer value from multiple dimensions, selecting variables that are highly discriminatory for customer value analysis, thereby improving the accuracy and reliability of customer value assessment. Compared to single-dimensional or simple assessment methods, this solution provides a more comprehensive understanding of customer behavior and value characteristics, providing banks with more targeted customer management strategies and enabling more efficient and reliable adjustments to customer credit limits.

[0021] Furthermore, the preset variable selection strategy is:

[0022] According to the corresponding categorical variables under a certain dimension, the information gain corresponding to each categorical variable under the dimension is calculated based on the preset information gain calculation formula;

[0023] The information gain calculation formula is:

[0024]

[0025] Where Gain(D,a) is the information gain of the categorical variable a relative to the dataset D in a certain dimension, the dataset D is the set of all categorical variables in this dimension, H(D) is the information entropy of the dataset D, and p i is the proportion of the i-th category variable in the data set D, k is the total number of categories of the category variable in the data set D, J is the number of values of the category variable a, D j Take the value a for the categorical variable a j The sample set, |D j | is the sample set D j The number of samples in , μ is the penalty coefficient, and C(a) is the complexity measure of the categorical variable a;

[0026] According to the information gain corresponding to each category variable, arrange them in order from high to low information gain to form the information gain arrangement table corresponding to this dimension;

[0027] According to the information gain ranking table corresponding to the dimension formed, based on the preset gain selection strategy, several category variables corresponding to the information gain are selected from the information gain ranking table in turn as the dimension evaluation variables corresponding to the dimension.

[0028] Beneficial Effects: This approach quantifies the classification power of each categorical variable for a dataset by calculating information gain. The greater the information gain, the more significant the variable's ability to distinguish between categories. Therefore, variable selection based on information gain can accurately identify key variables that significantly influence target dimensions (such as R, F, and M), avoiding the introduction of irrelevant or redundant variables, thereby improving the model's accuracy in customer value assessment and credit limit adjustments.

[0029] Real-world credit data can contain a significant amount of noise and irrelevant information. The penalty term μC(a) in this strategy penalizes highly complex variables, preventing them from being overly complex or prone to overfitting. Highly complex variables can contain significant noise and randomness, negatively impacting the model's generalization capabilities. The penalty term reduces the impact of noise on the model, making it more robust.

[0030] Furthermore, the preset customer value score calculation strategy is:

[0031] According to the dimension evaluation variables corresponding to each dimension, based on the preset dimension variable scoring table, determine the variable score corresponding to each dimension evaluation variable under each dimension;

[0032] According to the variable scores corresponding to the evaluation variables under each dimension, the customer value score corresponding to the customer is calculated based on the customer value score calculation formula;

[0033] The customer value score calculation formula is:

[0034]

[0035] In the formula, F is the customer value score, n, m, s are the total number of dimension evaluation variables corresponding to R dimension, F dimension and M dimension respectively, and f R(xi) is the variable score corresponding to the dimension evaluation variable xi under the R dimension, w R(xi) It is the weight value corresponding to the dimension evaluation variable xi under the R dimension.

[0036] Beneficial Effects: This solution calculates variable scores for the three dimensions of R (last borrowing date), F (borrowing amount), and M (borrowing frequency) and then weights them together to comprehensively reflect customer value from multiple perspectives. Variable scores across different dimensions capture diverse aspects of customer behavior. For example, R reflects customer activity, F reflects spending power, and M reflects transaction frequency. This comprehensive assessment avoids the limitations of single-dimensional evaluation, enabling banks to more accurately and comprehensively assess customer value. In actual lending operations, customer behavior patterns and value contributions vary. This strategy allows for flexible adjustment of weights across different dimensions, adapting to diverse customer needs and business scenarios. For example, for customers with small, frequent borrowing, the weights of F and M may be more important; whereas for customers with large, infrequent borrowing, the weights of R and F may be more important. By appropriately assigning weights, the value of different customer types can be more accurately assessed.

[0037] This customer value scoring calculation strategy features clear steps and formulas, forming a standardized process. Credit personnel can easily calculate customer value scores using the pre-defined dimensional variable scoring table and formula, reducing operational complexity and human error. This standardized process also facilitates unified management and oversight by banks, improving efficiency and consistency in business processing.

[0038] Furthermore, the customer value levels include high-value customer groups, medium-value customer groups, potential customer groups, and low-value customer groups.

[0039] Furthermore, the preset gain selection strategy is:

[0040] According to the information gain ranking table corresponding to the dimension formed, based on the preset gain threshold, the information gain greater than the preset gain threshold is selected, and the category variable corresponding to the selected information gain is matched as the dimension evaluation variable under the corresponding dimension; the size of the preset gain threshold is associated with the total number of information gains in the information gain ranking table.

[0041] Beneficial Effects: Among the many categorical variables, not all play a critical role in customer value assessment and credit limit adjustment. Using a preset gain threshold, we can filter out categorical variables with information gain greater than the threshold. These variables often contain information that is crucial to the target dimension (such as R, F, and M dimensions). For example, in the R dimension, this strategy can accurately identify variables that are closely related to the date of the customer's most recent loan and have a significant impact on the assessment. This eliminates the interference of irrelevant or redundant variables, allowing the model to focus more on core information, thereby improving its accuracy and effectiveness.

[0042] Too many variables can overcomplicate a model and lead to overfitting, where the model performs well on the training data but generalizes poorly to new data. Setting a preset gain threshold can control the number of variables entering the model, preventing overfitting caused by the model learning too much noise or unimportant information, thereby improving the model's stability and reliability in practical applications.

[0043] The preset gain threshold is linked to the total information gain in the information gain ranking table. This allows the threshold to be dynamically adjusted based on the actual data and business needs. When data distribution changes or business priorities shift, banks can adjust the gain threshold to reselect appropriate dimension evaluation variables, allowing the model to adapt promptly and maintain good performance.

[0044] Furthermore, the magnitude of the preset gain threshold is associated with the total number of information gains in the information gain arrangement table, including:

[0045] When the total information gain Q in the information gain arrangement table is greater than 0 and less than or equal to X, the preset gain threshold Y=1;

[0046] When the total information gain Q in the information gain arrangement table is greater than X and less than or equal to Y, the preset gain threshold Y=Q×T%; T is a preset percentage;

[0047] When the total information gain Q in the information gain arrangement table is greater than Y, the preset gain threshold Y=A×ln(Q)+B is used; A and B are dynamically adjustable constants.

[0048] Beneficial effects: In this solution, by setting the threshold calculation method corresponding to different ranges of total information gain, it can flexibly adapt to changes in data scale and characteristics. When the total information gain Q is in different intervals, different threshold calculation methods are used. For example, when Q is small (greater than 0 and less than or equal to X), the preset gain threshold is fixed to 1. This simple setting is suitable for situations where the amount of data is small and the overall level of information gain is not high, and it can quickly screen out relatively important variables. When Q is large, a more complex calculation method (such as Y = A × ln (Q) + B) is used. The threshold can be dynamically adjusted according to the richness of the data to adapt to more information and variable relationships in the data.

[0049] When the total information gain is small, a fixed threshold (Y = 1) can prevent excessive selection of unimportant variables due to a low threshold, which increases model complexity; it also prevents over-selection and loss of important information due to a high threshold. When the total information gain is large, using calculations related to Q (such as Y = Q × T% and Y = A × ln(Q) + B) can dynamically adjust the threshold based on the richness of the data, ensuring that the selected variables contain sufficient information to improve model accuracy without introducing too many redundant variables that may lead to model overfitting.

[0050] Different business scenarios may require different levels of strictness in variable screening. This association method allows for flexible adjustment of preset gain thresholds by adjusting parameters such as X, Y, T%, A, and B, based on business needs, to meet the selection requirements of dimension evaluation variables in different business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flowchart of a credit limit adjustment method based on RFM and decision tree model in Example 1 of the present invention. DETAILED DESCRIPTION

[0052] The following is further described in detail through specific implementation methods:

[0053] Example 1

[0054] A credit limit adjustment method based on RFM and decision tree model is basically as follows Figure 1 As shown, the following steps are included:

[0055] S1. Obtain the customer performance variable data corresponding to the R dimension, F dimension, and M dimension, process and organize the customer performance variable data corresponding to each dimension, and slice and classify them according to the preset slice categories to form category variables corresponding to each dimension; the R dimension, M dimension, and F dimension are the dimension of the time of the most recent loan, the dimension of the loan amount, and the dimension of the loan frequency, respectively; in this embodiment, slicing is performed according to multiple dimensions such as time and product to form variables corresponding to corresponding slice categories, such as the credit utilization rate in the past three months; the number of loans in the past six months, etc.

[0056] S2. Based on the preset variable selection strategy and the categorical variables corresponding to each dimension, several categorical variables with large information increments corresponding to the corresponding dimensions in each slice category are used as dimension evaluation variables for the corresponding dimensions;

[0057] The preset variable selection strategy is:

[0058] According to the corresponding categorical variables under a certain dimension, the information gain corresponding to each categorical variable under the dimension is calculated based on the preset information gain calculation formula;

[0059] The information gain calculation formula is:

[0060]

[0061] Where Gain(D,a) is the information gain of the categorical variable a relative to the dataset D in a certain dimension, the dataset D is the set of all categorical variables in this dimension, H(D) is the information entropy of the dataset D, and p i is the proportion of the i-th category variable in the data set D, k is the total number of categories of the category variable in the data set D, J is the number of values of the category variable a, D j Take the value a for the categorical variable a j The sample set, |D j | is the sample set D j The number of samples in , μ is the penalty coefficient, and C(a) is the complexity measure of the categorical variable a;

[0062] According to the information gain corresponding to each category variable, arrange them in order from high to low information gain to form the information gain arrangement table corresponding to this dimension;

[0063] According to the information gain ranking table corresponding to the dimension formed, based on the preset gain selection strategy, several category variables corresponding to the information gain are selected from the information gain ranking table in turn as the dimension evaluation variables corresponding to the dimension.

[0064] The preset gain selection strategy is:

[0065] According to the information gain ranking table corresponding to the dimension formed, based on the preset gain threshold, the information gain greater than the preset gain threshold is selected, and the category variable corresponding to the selected information gain is matched as the dimension evaluation variable under the corresponding dimension; the size of the preset gain threshold is associated with the total number of information gains in the information gain ranking table.

[0066] The magnitude of the preset gain threshold is associated with the total number of information gains in the information gain arrangement table, including:

[0067] When the total information gain Q in the information gain arrangement table is greater than 0 and less than or equal to X, the preset gain threshold Y=1;

[0068] When the total information gain Q in the information gain arrangement table is greater than X and less than or equal to Y, the preset gain threshold Y=Q×T%; T is a preset percentage;

[0069] When the total information gain Q in the information gain arrangement table is greater than Y, the preset gain threshold Y=A×ln(Q)+B is used; A and B are dynamically adjustable constants.

[0070] S3. Calculate the customer value score corresponding to each customer based on the dimensional evaluation variables corresponding to each dimension and the preset customer value score calculation strategy;

[0071] The preset customer value score calculation strategy is:

[0072] According to the dimensional evaluation variables corresponding to each dimension, based on the preset dimensional variable scoring table, the variable scores corresponding to each dimensional evaluation variable under each dimension are determined; in the preset dimensional variable scoring table in this embodiment, the quantile method is used to determine the importance of each dimensional variable segment, which is specifically divided into the upper quartile, the lower quartile, and the median, where the upper quartile is the most important (optimal value) segment, and the importance of the median and the lower quartile decreases in sequence. For example, a certain dimensional variable is arranged from small to large (assuming that the larger the value, the greater the contribution, and the higher the importance), the lower quartile is equal to the 25th percentile; the median is the 50th percentile; and the upper quartile is the 75th percentile. Then, the importance of the variable is less than or equal to the lower quartile, and the importance is greater than or equal to the upper quartile. Each segment corresponding to each dimensional variable has its own corresponding score, as shown in Table 1:

[0073]

[0074] Table 1

[0075] According to the variable scores corresponding to the evaluation variables under each dimension, the customer value score corresponding to the customer is calculated based on the customer value score calculation formula;

[0076] The customer value score calculation formula is:

[0077]

[0078] In the formula, F is the customer value score, n, m, s are the total number of dimension evaluation variables corresponding to R dimension, F dimension and M dimension respectively, and f R(xi) is the variable score corresponding to the dimension evaluation variable xi under the R dimension, w R(xi) It is the weight value corresponding to the dimension evaluation variable xi under the R dimension.

[0079] S4. Determine the customer value level for each customer based on the customer value score corresponding to each customer and a preset customer value score corresponding level table. The customer value levels include high-value customer groups, medium-value customer groups, potential customer groups, and low-value customer groups. In this embodiment, the customer value score corresponding level table uses an equal division method to divide customer value levels. For example, if the total score is E, the division corresponding to each customer value level is shown in Table 2:

[0080] Customer Ratings (0，E / 4] (E / 4, E / 2] (E / 2, 3E / 4] (3E / 4,E) Value Level Low-value customer groups potential customer base mid-value customer group High-value customer base

[0081] Table 2

[0082] S5. Determine the final credit limit level corresponding to each customer based on the determined customer value level corresponding to each customer and other customer indicator information corresponding to each customer.

[0083] In this embodiment, other indicator information of the customer includes the customer risk level or customer liability level corresponding to the customer.

[0084] For example, combine the customer value level with the customer risk level to determine the corresponding customer credit limit level.

[0085] High-value customers: High-value customers with a risk level of A enjoy the 7th level credit limit adjustment and obtain the most favorable credit limit adjustment plan; those with a risk level of B or C are in the 6th level; and those with a risk level of D are subject to the 5th level credit limit adjustment.

[0086] Medium-value customers: For medium-value customers with a risk level of A or B, the credit limit adjustment level is Level 6; when the risk level is C, the credit limit adjustment level is reduced to Level 5; when the risk level is D, the credit limit adjustment level is Level 4.

[0087] Potential customers: For potential customers with a risk level of A, the credit limit adjustment level is Level 5; for potential customers with a risk level of B, the credit limit adjustment level is Level 4; for potential customers with a risk level of C, the credit limit adjustment level is Level 3; and for potential customers with a risk level of D, the credit limit adjustment level is Level 2.

[0088] Low-value customers: For low-value customers with a risk level of A, the credit limit adjustment level is Level 4; when the risk level is B, the credit limit adjustment level is Level 3; when the risk level is C, the credit limit adjustment level is Level 2; when the risk level is D, only Level 1 credit limit adjustment can be obtained, which is also the most conservative credit limit adjustment arrangement.

[0089] S6. According to the final credit limit level corresponding to each customer and based on a preset credit limit adjustment table, determine the credit limit adjustment ratio corresponding to each customer under the corresponding final credit limit level.

[0090] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is excessively described here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A credit limit adjustment method based on RFM and decision tree model, characterized by: The following steps are involved: S1. Obtaining customer performance variable data corresponding to the R, F, and M dimensions, processing and organizing the customer performance variable data corresponding to each dimension, and slicing and classifying them according to preset slicing categories to form category variables corresponding to each dimension; the R, F, and M dimensions are the most recent loan time dimension, the loan frequency dimension, and the loan amount dimension, respectively; S2. Based on the preset variable selection strategy and the categorical variables corresponding to each dimension, several categorical variables with large information increments corresponding to the corresponding dimensions in each slice category are used as dimension evaluation variables for the corresponding dimensions; S3. Calculate the customer value score corresponding to each customer based on the dimensional evaluation variables corresponding to each dimension and the preset customer value score calculation strategy; S4. Determine the customer value level corresponding to each customer based on the customer value score corresponding to each customer and a preset customer value score corresponding level table; S5. Determine the final credit limit level for each customer based on the determined customer value level for each customer and other customer indicator information for each customer; S6. According to the final credit limit level corresponding to each customer and based on a preset credit limit adjustment table, determine the credit limit adjustment ratio corresponding to each customer under the corresponding final credit limit level.

2. The credit limit adjustment method based on RFM and decision tree model according to claim 1, characterized in that: The preset variable selection strategy is: According to the corresponding categorical variables under a certain dimension, the information gain corresponding to each categorical variable under the dimension is calculated based on the preset information gain calculation formula; The information gain calculation formula is: Where Gain(D,a) is the information gain of the categorical variable a relative to the dataset D in a certain dimension, the dataset D is the set of all categorical variables in this dimension, H(D) is the information entropy of the dataset D, and p i is the proportion of the i-th category variable in the data set D, k is the total number of categories of the category variable in the data set D, J is the number of values of the category variable a, D j Take the value a for the categorical variable a j The sample set, |D j | is the sample set D j The number of samples in , μ is the penalty coefficient, and C(a) is the complexity measure of the categorical variable a; According to the information gain corresponding to each category variable, arrange them in order from high to low information gain to form the information gain arrangement table corresponding to this dimension; According to the information gain ranking table corresponding to the dimension formed, based on the preset gain selection strategy, several category variables corresponding to the information gain are selected from the information gain ranking table in turn as the dimension evaluation variables corresponding to the dimension.

3. The credit limit adjustment method based on RFM and decision tree model according to claim 2, characterized in that: The preset customer value score calculation strategy is: According to the dimension evaluation variables corresponding to each dimension, based on the preset dimension variable scoring table, determine the variable score corresponding to each dimension evaluation variable under each dimension; According to the variable scores corresponding to the evaluation variables under each dimension, the customer value score corresponding to the customer is calculated based on the customer value score calculation formula; The customer value score calculation formula is: In the formula, F is the customer value score, n, m, s are the total number of dimension evaluation variables corresponding to R dimension, F dimension and M dimension respectively, and f R(xi) is the variable score corresponding to the dimension evaluation variable xi under the R dimension, w R(xi) It is the weight value corresponding to the dimension evaluation variable xi under the R dimension.

4. The credit limit adjustment method based on RFM and decision tree model according to claim 3, characterized in that: The customer value levels include high-value customer groups, medium-value customer groups, potential customer groups, and low-value customer groups.

5. The credit limit adjustment method based on RFM and decision tree model according to claim 4, characterized in that: The preset gain selection strategy is: According to the information gain ranking table corresponding to the dimension formed, based on the preset gain threshold, the information gain greater than the preset gain threshold is selected, and the category variable corresponding to the selected information gain is matched as the dimension evaluation variable under the corresponding dimension; the size of the preset gain threshold is associated with the total number of information gains in the information gain ranking table.

6. The credit limit adjustment method based on RFM and decision tree model according to claim 5, characterized in that: The magnitude of the preset gain threshold is associated with the total number of information gains in the information gain arrangement table, including: When the total information gain Q in the information gain arrangement table is greater than 0 and less than or equal to X, the preset gain threshold Y=1; When the total information gain Q in the information gain arrangement table is greater than X and less than or equal to Y, the preset gain threshold Y=Q×T%; T is a preset percentage; When the total information gain Q in the information gain arrangement table is greater than Y, the preset gain threshold Y=A×ln(Q)+B is used; A and B are dynamically adjustable constants.

Images