Enhancing Loan Approval Prediction: A Comparative Study of Machine Learning, Deep Learning, and Bayesian Networks

Accurate assessment of loan application risk is critical in the financial sector to minimize defaults and mitigate financial losses. Financial institutions leverage advanced computational techniques, including Machine Learning (ML), Deep Learning (DL), and Bayesian Networks, to enhance risk evaluation and decision-making processes. While models such as Random Forest (RF) offer superior predictive accuracy, their lack of interpretability poses challenges in regulatory compliance. Conversely, Bayesian Networks provide a more transparent decision-making framework, enabling the assessment of key borrower attributes such as credit score, income, and loan amount. The study evaluates the effectiveness of Machine Learning, Deep Learning and Bayesian Networks in loan approval prediction using real-world datasets. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is employed, while Mutual Information is utilized for feature selection. Experimental results demonstrate that Random Forest (RF) achieves the highest predictive accuracy, whereas Bayesian Networks offer improved interpretability while maintaining competitive performance. Despite their accuracy, deep learning models exhibit limited explainability. The findings highlight the trade-offs between model accuracy and interpretability, providing insights for financial institutions to adopt AI-driven loan approval models that align with both predictive performance and regulatory requirements.