A Fuzzy C-Means and Sugeno Fuzzy Inference System for Credit Card Default Prediction Using Mutual Information-Based Feature Selection

Accurate prediction of credit card default is a critical task in financial risk management, where timely identification of high-risk clients directly affects the stability of lending institutions. This study proposes an interpretable hybrid framework that combines Fuzzy C-Means (FCM) clustering with a Sugeno-type Fuzzy Inference System for credit card default prediction. To improve model efficiency and reduce input dimensionality, a Mutual Information-based feature selection strategy is employed to retain the most informative attributes. In the proposed framework, FCM partitions the input space into overlapping clusters, and the resulting membership degrees are used to construct fuzzy rules within the Sugeno inference system. This design enables the model to capture nonlinear and uncertain relationships in financial data while maintaining an interpretable decision-making structure.

The proposed framework was evaluated on the benchmark Default of Credit Card Clients dataset and compared with Logistic Regression and Random Forest baselines. Experimental results show that the proposed model achieved an accuracy of 77.53%, a recall of 53.73%, an F1-score of 51.41%, and an ROC-AUC of 68.52%. The proposed approach outperformed both baseline models in terms of recall and F1-score, demonstrating improved detection capability for default cases. Although Random Forest achieved higher overall accuracy, McNemar’s test confirmed statistically significant differences in prediction behavior between the proposed framework and the baseline classifiers, highlighting the distinct decision-making characteristics of the fuzzy model.