Rejection Inference Based On Label Propagation Is Realized By Combining Transfer Learning And Semi-supervised Learning

With the development of artificial intelligence technology,all industries have been deeply affected or even changed dramatically.In the financial field,it is mainly about the boosting of financial-technology.Credit risk control is an important part of fintech.At present,the common credit The risk control method is based on the existing data and combines data mining and machine learning methods from a quantitative perspective to build models.It specifically includes machine learning methods such as logistic regression,decision trees,and deep learning methods such as neural networks.However,there is a big problem in the modeling process: the sample data we have are all“goodly rated”passing users,and there are no “badly rated”rejected customers.In the long run,the model is for the overall sample There will be a bias,because the test sample facing the model is an overall sample,and our training sample only has data for users with“good scores”and no users with“bad scores”,so the trained model is in“good scores”users are getting higher and higher scores,but ” good scores ” users can't get any verification.With the continuous in-depth research of the academia and industry,relevant scholars have proposed methods to solve this problem from different angles,which we called rejecting inference.This paper uses a combination of transfer learning and semi-supervised learning ideas and applies it to the rejection inference algorithm.It makes full use of unlabeled data to implement the rejection inference based on label propagation,so as to better infer the credit performance of rejected customers.By solving the problem of sample bias in the credit scoring problem,we can obtain a relatively complete modeling sample,which makes our credit scoring more accurate.