Deep Learning-Based Credit Risk Prediction Using Convolutional Neural Networks

Credit scoring has always been conceived as an important element to financial risk evaluation enabling institutions to make informed lending decisions. Conventional credit scoring employs historical financial information and rule-based models that tend not to capture non-linear trends in consumer behavior. Machine learning and other models can improve risk prediction through an empirical approach. In this paper, we examine the performance of a CNN derived approach for credit behaviour prediction using the Kaggle Credit Card Behaviour Score dataset. The ROC AUC score of the model is 1.00, indicating a very good classification. Results suggest that CNNs provide an opportunity to enhance credit risk prediction by discovering novel financial patterns. Challenges of overfitting and class imbalance need further tuning for implementation in financial decision-making.