A Study On The Application And Evaluation Of Machine Learning Algorithms In Personal Loan Default Prediction

The management of credit risk has always been a hot issue of academic concern,with the development of artificial intelligence and financial technology,machine learning algorithms have gradually replaced traditional risk control models,especially in the field of personal loan default risk prediction,a large number of studies have shown that machine learning algorithms have higher accuracy and wider applicability.However,the vast majority of the literature uses an evaluation index based on a confusion matrix when evaluating the performance of the model,and less on the characteristics of personal credit business.Therefore,this paper supplements the evaluation indicators combined with the characteristics of personal credit business,and further compares the effect of the analysis model,which can not only help financial institutions choose the best algorithm,but also provide basis and empirical support for the expansion of the evaluation index system of machine learning.This paper uses multiple public personal credit data sets,applies ten machine learning algorithms for default prediction,and builds a personal credit default prediction model.A comparative analysis of personal credit default prediction models is carried out from several aspects.Through empirical analysis,this paper finds that the Boosting ensemble algorithm is better than other types of algorithms.Among the Boosting algorithms,the Cat Boost algorithm performs the best.In addition,the traditional classification ability evaluation indicators have certain shortcomings—that is,they cannot fully describe the performance of the model in terms of improvement effect and predicted income,and the relative improvement and expected income evaluation indicators supplemented in this paper can make up for this to a certain extent.Based on the empirical results,this paper finds that the Cat Boost algorithm has relatively good performance in the traditional classification ability evaluation index,relative improvement class and expected return evaluation index.Financial institutions can use the Cat Boost algorithm as the current benchmark algorithm for personal credit default prediction.Relative improvement category,expected income category and other evaluation indicators combined with the characteristics of personal loan business,and update and improve according to their own data and business logic.