Modeling method for risk assessment in overdue state

Abstract

The invention discloses a modeling method for risk assessment in an overdue state, and the method comprises the following steps: S1), constructing an enforcement model, specifically comprising the following steps: S11), defining an enforcement label; s12) collecting historical data of credit customers; s13) determining a threshold value and a weight of the label; s14) performing model training; s2) constructing a recovery model; the method specifically comprises the following steps: S21) defining a recycling label; s22) collecting historical data of the urged customer; s23) determining a threshold value and a weight of the label; s24), performing model training; s3) performing cross use and strategy optimization; the method specifically comprises the following steps: S31) drawing a customer portrait; s32), making a personalized strategy; and S33) continuous monitoring and timely feedback are carried out. According to the invention, the accuracy of credit risk assessment is improved, and the credit loss is effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the field of credit technology, and in particular relates to a modeling method for risk assessment in an overdue state. Background Art

[0002] Traditional credit risk assessment models often use a relatively crude label classification method, such as only distinguishing between "overdue" and "not overdue", and failing to deeply explore the multi-dimensional information behind overdue. This simplified risk assessment method easily leads to inaccurate judgment of credit risk and fails to accurately identify the differentiated risks of different customers. Summary of the invention

[0003] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0004] A modeling method for risk assessment in an overdue state includes the following steps:

[0005] S1): Build an entry reminder model: Train the entry reminder scoring model by defining multiple label parameters;

[0006] The specific steps are as follows:

[0007] S11): Define reminder labels: Define outgoing labels including overdue days, overdue amount, overdue times, and overdue type;

[0008] S12): Collect historical data of credit customers: Collect historical data, including credit information, borrowing information, repayment information, and credit scores;

[0009] S13): Determine the threshold and weight of the tag: According to the characteristics of the financial product and the historical data analysis of the customer, determine the threshold and weight of each tag;

[0010] S14): Model training: Use machine learning to train the incoming and outgoing reminder labels and build an incoming and outgoing reminder model;

[0011] S2): Building a recycling model: training a recycling scoring model by defining multiple label parameters;

[0012] The specific steps are as follows:

[0013] S21): Define recycling labels: Define labels including recycling amount, recycling time, collection method, product attributes and customer characteristics;

[0014] S22): Collect historical data of customers who are in debt collection: Collect historical data, including credit information, loan information, repayment information, and credit score;

[0015] S23): Determine the threshold and weight of the tag: Determine the threshold and weight of each tag based on the customer's historical data analysis;

[0016] S24): Model training: Use machine learning to train recycling labels and build a recycling model;

[0017] S3): Cross-use and strategy optimization: Combine the scoring data of the two models and give a comprehensive score;

[0018] The specific steps are as follows:

[0019] S31): Customer profile: According to the scores of the collection model and the collection model, customers are classified into low collection and low collection, low collection and high collection, high collection and low collection, and high collection and high collection;

[0020] S32): Personalized strategy formulation: Based on different categories of customer portraits, formulate refined credit approval strategies and collection strategies;

[0021] S33): Continuous monitoring and timely feedback: Based on the model's prediction scores, establish a feedback mechanism to monitor changes in customer credit risk in real time.

[0022] Furthermore, in step S11, the overdue type is the first overdue or multiple overdue.

[0023] Furthermore, in step 14 and step 24, the machine learning algorithm may be logistic regression, random forest or GBDT.

[0024] Furthermore, the scores of the scoring models in step 1 and step 2 range from 0 to 100;

[0025] When the urging model score is <50, it is low urging;

[0026] When the score of the urging model is >= 50, it is a high urging;

[0027] When the recovery model score is <50, it is low recovery;

[0028] When the recovery model score is >= 50, it is high recovery.

[0029] Furthermore, the comprehensive scoring rule in step 3 is: 40% for collection + 60% for recovery.

[0030] Beneficial effects of the present invention:

[0031] The present invention evaluates the customer's overdue risk in a more refined manner by disassembling overdue labels, such as collection and recovery, which helps financial institutions to more effectively identify high-risk customers such as high collection and low recovery and low-risk customers such as low collection and high recovery, providing strong support for subsequent risk control, improving the accuracy of credit risk assessment, and effectively reducing credit losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a schematic diagram of the system provided by the present invention; DETAILED DESCRIPTION

[0034] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention. Specific embodiment 1:

[0036] like Figure 1 As shown, a modeling method for risk assessment under overdue status includes the following steps:

[0037] S1): Build an entry reminder model: Train the entry reminder scoring model by defining multiple label parameters;

[0038] The specific steps are as follows:

[0039] S11): Define reminder labels: Define labels including overdue days, overdue amount, number of overdue times, and overdue type; the overdue days here can be based on a cycle of five days, the overdue amount can be based on a cycle of ten thousand, and the number of overdue times can be based on a cycle of 1 time. The lower the score is in each next cycle, the overdue type is the first overdue or multiple overdue.

[0040] S12): Collect historical data of credit customers: Collect historical data, including credit information, borrowing information, repayment information, and credit scores;

[0041] S13): Determine the threshold and weight of the tag: According to the characteristics of the financial product and the historical data analysis of the customer, determine the threshold and weight of each tag; the threshold is set according to the characteristics of different financial products, and the weight is set according to the historical data of the customer.

[0042] S14): Model training: Use machine learning to train the incoming and outgoing reminder labels and build an incoming and outgoing reminder model;

[0043] S2): Building a recycling model: training a recycling scoring model by defining multiple label parameters;

[0044] The specific steps are as follows:

[0045] S21): Define recycling labels: Define labels including recycling amount, recycling time, collection method, product attributes and customer characteristics;

[0046] S22): Collect historical data of customers who are in debt collection: Collect historical data, including credit information, loan information, repayment information, and credit score;

[0047] S23): Determine the threshold and weight of the tag: Determine the threshold and weight of each tag based on the customer's historical data analysis;

[0048] S24): Model training: Use machine learning to train recycling labels and build a recycling model;

[0049] The scores of the scoring models in step 1 and step 2 range from 0 to 100;

[0050] When the urging model score is <50, it is low urging;

[0051] When the score of the urging model is >= 50, it is a high urging;

[0052] When the recovery model score is <50, it is low recovery;

[0053] When the recovery model score is >= 50, it is high recovery.

[0054] In step 14 and step 24, the machine learning algorithm may be logistic regression, random forest or GBDT.

[0055] S3): Cross-use and strategy optimization: Combine the scoring data of the two models and make a comprehensive score; the comprehensive scoring rule in step 3 is: 40% for collection + 60% for recovery.

[0056] The specific steps are as follows:

[0057] S31): Customer profile: According to the scores of the collection model and the collection model, customers are classified into low collection and low collection, low collection and high collection, high collection and low collection, and high collection and high collection;

[0058] S32): Personalized strategy formulation: Based on different categories of customer portraits, formulate refined credit approval strategies and collection strategies;

[0059] S33): Continuous monitoring and timely feedback: Based on the model's prediction scores, establish a feedback mechanism to monitor changes in customer credit risk in real time.

[0060] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A modeling method for risk assessment in overdue state, characterized by: The following steps are involved: S1): Build an entry reminder model: Train the entry reminder scoring model by defining multiple label parameters; The specific steps are as follows: S11): Define reminder labels: Define outgoing labels including overdue days, overdue amount, overdue times, and overdue type; S12): Collect historical data of credit customers: Collect historical data, including credit information, borrowing information, repayment information, and credit scores; S13): Determine the threshold and weight of the tag: According to the characteristics of the financial product and the historical data analysis of the customer, determine the threshold and weight of each tag; S14): Model training: Use machine learning to train the incoming and outgoing reminder labels and build an incoming and outgoing reminder model; S2): Building a recycling model: training a recycling scoring model by defining multiple label parameters; The specific steps are as follows: S21): Define recycling labels: Define labels including recycling amount, recycling time, collection method, product attributes and customer characteristics; S22): Collect historical data of customers who are in debt collection: Collect historical data, including credit information, loan information, repayment information, and credit score; S23): Determine the threshold and weight of the tag: Determine the threshold and weight of each tag based on the customer's historical data analysis; S24): Model training: Use machine learning to train recycling labels and build a recycling model; S3): Cross-use and strategy optimization: Combine the scoring data of the two models and give a comprehensive score; The specific steps are as follows: S31): Customer profile: According to the scores of the collection model and the collection model, customers are classified into low collection and low collection, low collection and high collection, high collection and low collection, and high collection and high collection; S32): Personalized strategy formulation: Based on different categories of customer portraits, formulate refined credit approval strategies and collection strategies; S33): Continuous monitoring and timely feedback: Based on the model's prediction scores, establish a feedback mechanism to monitor changes in customer credit risk in real time.

2. A modeling method for risk assessment under overdue status according to claim 1, characterized in that: In the step S11, the overdue type is the first overdue or multiple overdue.

3. The modeling method for risk assessment under overdue status according to claim 1 is characterized by: In step 14 and step 24, the machine learning algorithm may be logistic regression, random forest or GBDT.

4. The modeling method for risk assessment under overdue status according to claim 1 is characterized by: The scores of the scoring models in step 1 and step 2 range from 0 to 100; When the urging model score is <50, it is low urging; When the score of the urging model is >= 50, it is a high urging; When the recovery model score is <50, it is low recovery; When the recovery model score is >= 50, it is high recovery.

5. The modeling method for risk assessment under overdue status according to claim 1 is characterized by: The comprehensive scoring rule in step 3 is: 40% for collection + 60% for recovery.

Images