FusionXAI: A Transparent Ensemble Framework for Credit Card Fraud Prediction
Students & Supervisors
Student Authors
Supervisors
Abstract
Credit card fraud is a crucial problem in financial security that requires accurate but also interpretable and trustworthy predictive systems. It proposes a new ensemble-based fraud prediction framework called FusionXAI, which combines high-performance machine learning classifiers, including Random Forest, LightGBM, XGBoost, and Neural Networks with explainable AI (XAI) techniques to provide model performance and interpretability. To handle the extreme class imbalance nature of real-world transaction datasets, the framework uses hybrid data-balancing strategies adopted from the Synthetic Minority Over-sampling Technique (SMOTE) and random under sampling approach. FusionXAI contributes to improved robustness and accuracy by controlling the complementary elements of diversity in large groups of learners, thereby reducing false positives and providing flexibility to changing fraud environments. Moreover, Using SHAP and LIME helps explain how a model makes decisions, making the results easier to check and trust. Experimental results on a real-world credit card transaction dataset show that the proposed FusionXAI framework achieves an F1-score of 0.8362, a recall of 0.7789, and an AUC-ROC of 0.9425. The framework performs better than individual models like Random Forest, which has an F1-score of 0.8276, and it performs similarly to XGBoost, which has an AUC-ROC of 0.9751. FusionXAI provides accurate fraud detection with fewer false alarms, making it a reliable and scalable system for modern credit card fraud detection.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 5th International Conference on Electrical, Computer, & Telecommunication Engineering (ICECTE 2026)
- Date of Conference: 29/01/2026 - 29/01/2026
- Venue: Rajshahi, Bangladesh.
- Organizer: Faculty of Electrical & Computer Engineering (ECE), RUET