Research On Risk Evaluation Of P2P Online Loan Platform Based On Machine Learning

Peer-to-peer lending platforms have been developing rapidly in recent years.But a variety of adverse phenomena have emerged,the number of bad platforms is increasing,which has greatly affected the development of the online lending industry and people's investment enthusiasm.Great concerns should be paid attention to the risk of the platform.Therefore,how to evaluate the risk of online lending platform reasonably and accurately is an urgent problem to be solved.This topic mainly focuses on this problem.This paper focuses on P2 P online lending platform for data analysis.On the one hand,the user's attitude towards the platform,the user's focus and advantages and disadvantages of the platform are observed through the sentiment analysis and topic analysis of the comments on the online lending platform.On the other hand,we collect the index data disclosed by the online lending platform on relevant websites and select the characteristics of each explanatory variable,and then establish a variety of classification models on the data set by using machine learning related algorithms.Through model comparison,the optimal model is selected to help customers determine whether the platform is a normal platform or a problem platform,so as to prevent the risk of the platform.Through the analysis two conclusions are drawn: Firstly,it is found that most users have positive emotions towards the platform,and the Latent Dirichlet Al ocation thematic model is used to make people have a further understanding of the advantages and disadvantages of the online lending platform.Secondly,it is found that the prediction effect of Gradient Boosting Decision Tree model is relatively better,which can be used for assessing platform risk.The innovation of this paper mainly lies in the following three aspects.Firstly,a new research method named text analysis is widely adopted to dig into user's emotions and concerns through unstructured information,that is,the content of user's comments.Secondly,the feature selection based on filtering method is adopted to determine the research variables,and the variables with high correlation with the target variables are screened out.Thirdly,a variety of new classification models such as K-Nearest Neighbor,Naive Bayes Model and Gradient Boosting Decision Tree model are used for comparative analysis,besides the common-used classification models,such as Random Forest and Support Vector Machine.It is concluded that the prediction result of Gradient Boosting Decision Tree model is better and the prediction accuracy is 78.1%,which is better than the previous prediction results.