Loan defaults are important to correctly forecast, so as to ensure the survival and profitability of financial institutions. Thecommonly used traditional credit scoring models do not always reflect nonlinear connections between borrower behaviorand macroeconomic circumstances, which are rather complex. This paper examines the relative efficacy of ensemblemachine learning models and state-of-the-art AI-based credit scoring systems when it comes to predicting loan defaults.We train, compare and test various models: Random Forest, Gradient Boosting, and deep learning-based hybrids usingan actual dataset of lending. These results prove that ensemble approaches are much more effective in predicting theoutcome compared to older models and AI-based models using alternative data sources offer an even higher potential ofrisk assessments. These findings support the benefits of explainable AI approaches as a way to strike the balance betweeninterpretability and performance on the one hand and provide real-world advising to the lending sector, regulatoryauthorities, and fintech start-ups on how to streamline the management of credit risk.
Olatoye Kabiru Agboola (Thu,) studied this question.