Predicting potential defaulters of Internet consumer loan products based on improved logistic regression and RBF neural network algorithm

This study aims to enhance risk management in internet-based consumer loan platforms by predicting potential defaulters using advanced machine learning algorithms and conducting comprehensive business analysis. We analyze various factors influencing loan default risks, leveraging a self-built dataset from a Chinese internet loan platform. The dataset includes features such as Income, Age, Experience, Single_married, House_Ownership, Car_Ownership, Current_Job_Yrs, Current_House_Yrs, Health_Status. We employ logistic regression and radial basis function (RBF) neural network models to develop predictive tools, comparing their performance based on accuracy and AUC values. The logistic regression model achieved an accuracy of 89.10% and an AUC value of 0.75, while the RBF neural network model showed an accuracy of 90.40% and an AUC value of 0.79. Combining logistic regression with the RBF neural network improved accuracy to 91.10% and AUC value to 0.80.