Research And Predoction Of Online Lending Behavior Based On Muti-feature Information Fusion

With the rise of Internet finance and the concept of inclusive finance,online loans have become an important part of Internet finance with their low threshold,flexible operation and high returns.Since the launch of China’s first P2P platform in 2007,it has provided new funding channels for many small and micro enterprises and individuals,and also provided a new way of financial management.However,with the continuous development of the industry,the continuous competition in the industry,the imperfect credit reporting system and the long-term lack of supervision have made the industry risk accumulate continuously,which caused the withdrawal of the entire industry.In this study,the data of Renrendai are selected to study the influencing factors of investors’ investment behavior and borrowers’default behavior based on multi-feature information,and a variety of machine learning methods are used to predict.This paper identifies the text information including loan title and loan description from three perspectives of text characteristics,text content and text emotion,and analyzes the correlation between such information and loan behavior by using Logit regression model.At the same time,Catboost model and other machine models are used to predict online lending behavior by using multiple information including text information,and the prediction effects of different machine learning models are evaluated.The empirical results show that:(1)Most of the text characteristics,text sentiment,and text content have a positive impact on investors’investment,and the richer and more comprehensive the information reflected in the text,the easier it is to get a loan.(2)Different text information has different influences on the default rate of borrowers.Borrowers who use more strong words are more likely to default.The more information mentioned in the text,the lower the default rate of borrowers.(3)Catboost model has a better performance in predicting investors’investment behavior and borrowers’ default behavior than other models because it can directly deal with classification variables.Through comparative research methods and empirical research methods,this paper summarizes the reasons for the withdrawal of the P2P industry,and provides certain reference significance for the use of multiple information to further improve credit assessment and credit risk identification.