Loan Default Prediction Based on Machine Learning Approaches

To address the credit risk losses incurred by commercial banks due to loan defaults, this study utilizes the loan default prediction dataset from the Alibaba Tianchi platform to develop machine learning models for predicting customer defaults, aiming to mitigate credit risk. Given the characteristics of class imbalance and high dimensionality of loan data, data preprocessing and exploratory data analysis are conducted. Based on a comparative analysis of various models, seven machine learning algorithms that demonstrate superior performance are selected for experimental comparison, including Decision Tree, Random Forest, AdaBoost, Bagging, XGBoost, LightGBM, and CatBoost. The results indicate that ensemble learning algorithms exhibit higher accuracy and predictive performance compared to single algorithms, with the CatBoost model performing best across various indicators, including AUC. The study identifies key features highly correlated with loan defaults, including loan grade, annual income, loan amount, credit history length, and debt-to-income ratio.