Abstract

Credit card default is a major risk in banking, causing significant financial losses and instability. This study evaluates and combines three classification models Extreme Gradient Boosting (XGBoost), Artificial Neural Network), and Fuzzy using the UCI Credit Card dataset with 30,000 records and 25 features. The data were preprocessed through normalization, irrelevant feature removal, and split into 80% training and 20% testing. Model performance was assessed with Accuracy, Precision, Recall, F1-score, and AUC. Results show that XGBoost achieved the highest accuracy (95.61%, AUC = 0.95), Fuzzy obtained the highest recall (0.94), while ANN reached perfect precision (1.00). The hybrid Artificial Neural Networks-Xgboost model provided a balanced trade-off between precision and recall, whereas Artificial Neural Networks-Fuzzy performed the weakest. These findings highlight that XGBoost is the most effective model for accuracy, Fuzzy is more sensitive in detecting defaults, and Artificial Neural Networks-Xgboost offers a practical balance for credit risk management. Future improvements may involve data balancing techniques such as SMOTE to enhance sensitivity to minority cases.

Concepts :

Access to Document

Citations by Year