An Ensemble KAN-XGBoost Model for Fraud Detection

Financial fraud represents a growing challenge for financial institutions and e-commerce, requiring increasingly sophisticated detection methods. Traditional machine learning models, while effective, can reach limitations when facing complex fraud patterns and highly imbalanced datasets. This paper proposes a novel ensemble approach, KAN-XGBoost, which combines the power of Kolmogorov-Arnold Networks (KAN) for learning complex relationships with the robustness of the Extreme Gradient Boosting (XGBoost) algorithm for high-performance classification. Using the synthetic PaySim dataset, we demonstrate the effectiveness of our approach. To address the severe class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied to the training data. Our experimental results show that the KAN-XGBoost ensemble model, in soft voting configuration, significantly outperforms the individual models, achieving a performance metrics of 99%. This high performance suggests that the hybridization of KANs with established boosting algorithms constitutes a promising avenue for enhancing the security of financial transactions.