Bridging Tradition and Innovation: A Comparative Analysis of Credit Risk Models

This research compares the effectiveness of deep neural networks (DNNs) with traditional and contemporary machine learning techniques. A Taiwanese credit card transaction dataset serves as the basis for this research which originated from the UCI Machine Learning Repository. The dataset contains information about 30,000 users representing those who defaulted payment (6,636) and customers who kept up with payments (23,364). The research team assessed various models starting with Logistic Regression through Decision Trees and then Random Forests followed by Support Vector Machine (SVM) and finally Deep Neural Networks (DNN). A DNN generated superior predictive accuracy at 81.91% compared to alternative models included in this study. This study demonstrates that both feature scaling techniques along with hyperparameter adjustment led to enhanced DNN performance in predicting credit default behav-iors. Findings from this research provide real-world guidance for financial institutions that aim to upgrade their credit risk management systems through more effective decision-making processes.