Default Risk Prediction Using Random Forest and XGBoosting Classifier

Since the prosperity of internet financial services, micro-lending has become popular in recent years. To prevent the default risk and to predict the risk in advance, many traditional methods have been applied such as credit scoring, Logistic Regression (LR), Bayesian Decision Rules. To enhance the prediction accuracy, this study incorporates sequential feature selection with machine learning models, Random Forest and XGBoosting Classifier, to predict default risk based on a peer-to-peer lending platform. The LendingClub data has been used for analysis. Our experimental results show that Random Forest and XGBoosting Classifier both have an accuracy as 0.97, but if we take the ROC into account then the Random Forest model performs better.