Enhancing Credit Risk Prediction Using Deep Learning Techniques

Abstract

Accurate credit risk prediction is crucial for financial institutions seeking to minimize loan defaults as well as make more informed decisions. The study proposes the utilization of a deep learning-based approach for loan applicants to be classified as creditworthy or non-creditworthy by deploying a multilayer perceptron (MLP) neural network. Demographic characteristics and some
credit-related characteristics are utilized as input in the proposed model. To address the class imbalance problem common in credit data, SMOTEENN (Synthetic Minority Oversampling Technique–Edited Nearest Neighbours) is implemented. The quality of models is tested utilizing the major classification metrics: precision, recall, F1-score, and confusion matrix assessment. The results validate
that the proposed MLP architecture effectively identifies patterns in consumer credit behavior and offers excellent predictive power, even when confronted with unbalanced data. This paper supports the potential of deep neural networks to be promising tools for enhancing credit risk evaluation in modern banking systems.