Overdue risk identification method based on robust algorithm

Abstract

The invention discloses an overdue risk identification method based on a robust algorithm. The method comprises the steps: obtaining basic information of a user and multi-dimensional behavior data, and generating basic information behavior data ; and modeling the basic information behavior data by using a loss function of a robust algorithm to generate model data. According to the method, the problem that in the prior art, independent identical distribution conditions are not met due to the fact that sample distribution is damaged can be solved, and the recognition accuracy and robustness canbe improved. Compared with the prior art, the method has the following advantages and positive technical effects: 1, the anti-interference performance and robustness of the algorithm are effectively improved by using the minimum and maximum game theory; 2, compared with a traditional method based on down-sampling or up-sampling, the model has higher precision; and 3, original sample distribution is well described, and the situation of sample imbalance or sample non-independent same distribution can be effectively handled;

Description

technical field [0001] The invention relates to a method for identifying the overdue risk of consumer loans, in particular to an overdue risk identification method based on a robust algorithm. Background technique [0002] Consumer loans are also known as "consumer loans". Loans extended to consumers personally to purchase durable consumer goods or pay various expenses. [0003] Risk control is the lifeline of the credit sector. Since 2019, there have been obvious signs of an increase in the non-performing rate of various credit institutions, and risk control has also received unprecedented attention. [0004] The expansion of consumer credit scale is closely related to the development of intelligent risk control. Intelligent risk control refers to the multi-dimensional behavior data of application users to predict whether they will repay their loans on time. [0005] Usually, the overdue ratio of consumer loan customers is less than 5%, which is a scenario where positiv...

AI Technical Summary

Problems solved by technology

But it will inevitably lead to independent and identically distributed (independent and identically distributed) conditions are no longer satisfied, so the classic training algorithm will no longer be applicable

Images