Machine Learning-Based Default Loan Prediction for Financial Risk Assessment in Digital Lending

Abstract

In recent decades, the global financial system has undergone tremendous change. While capital markets play a crucial role in some situations and traditional banks and financial institutions continue to be the main source of funding for businesses and people in the majority of nations, new digital lending platforms have lately emerged. In this paper, the author introduces a viable machine learning-based approach to the estimation of financial risk in online lending on the Lending Club dataset. The suggested solution starts with an extensive data collection and exploratory analysis, followed by intensive preprocessing of data that includes outlier treatment, feature engineering, one-hot encoding, and normalization, as well as the class imbalance mitigation method of SMOTE. A systematic hyperparameter optimization is applied to creating an optimized Extreme Gradient Boosting (XGBoost) model that is used to ensure that nonlinear relationships between borrower and loan traits are learned. There are various performance measures that are used to evaluate the model, some of them include accuracy, precision, recall, F1-score, ROC curve, and confusion matrix analysis. The experimental outcomes are highly predictive with an accuracy of 97.70, precision of 94.78, recall of 93.30 and a near perfection of the AUC of 0.9991. The fact that the proposed model is generalizable compared to the traditional machine learning models also justifies the superiority of the given approach and shows that it has power, can be scaled, and placed in the context of applying it into practice to evaluate financial risks in the digital lending model.