Anti-fraud model with repayment recognition capability as core

Abstract

The invention discloses an anti-fraud model taking repayment recognition capability as a core, and solves the technical problems that the current model is not rigorous enough, the risk recognition is not perfect enough and the accuracy is low. Reading an identity card number of a customer to be loan, and initializing a score R of the customer to be checked to be equal to 0; taking the identity card number as an index to extract corresponding data of a related database; if the identity card number hits fields related to social stability and the like, directly returning a negative result, and if not, entering; if the identity card number hits the related feature 4, each R + 20 is hit; if the related feature 5 is hit, hit each R + 15; obtaining an accumulated R total score in the step; if Rgt; if not, returning a negative result; if 0 lt; rlt; 60, if yes, a judicial result is returned, and the financial institution is suggested to further carefully check; and if R is equal to 0, returning a continuation result. According to the method, a logic trunk is sorted out from a plurality of influence factors to identify a real risk source, and the logic trunk can be used as a footstone and preposed input of an existing anti-fraud model system.

Description

technical field [0001] The invention relates to the technical field of financial anti-fraud, in particular to an anti-fraud model centered on identifying repayment ability. Background technique [0002] The more common loan review by financial institutions is to conduct credit evaluation based on the authorization information of loan applicants, and establish a static or dynamic rule base based on manual experience based on the combination of experience and credit evaluation data, and use mainly from the People's Bank of China. Asset information, known blacklist database and other data are used to establish a rule-based credit risk model. When the model output is not good or hits the business risk code, the loan application is rejected. Another machine learning model represented by the scorecard model is mainly based on the construction of many features and multiple data sources as training sets, verification sets and test sets. The models include random forest, XGBoost, gra...

AI Technical Summary

Problems solved by technology

[0003] The existing rule model does not focus on the rapid loss of repayment ability, lacks a deep understanding of financial fraud risk control, and has not established a rigorous and effective risk identification system. It mainly piles up rules refined from scattered experience, resulting in hits The established rules can detect it, but in the face of ever-changing fraud methods, most models cannot effectively deal with it

[0004] The model based on machine learning methods relies on a large amount of user data and third-party data. Not only is the legitimacy of many third-party data problematic, the continuity of the business cannot be guaranteed, and due to the lack of a sufficient number of fraud-related samples, especially In the face of new fraud problems, the samples are unbalanced during model training, the accuracy and recall rate of prediction cannot meet the actual needs, and the generalization ability is not good when dealing with different financial business scenarios.

In addition, the current machine learning model is mainly to identify credit risk issues

Images