Kolmogorov-Arnold Neural Networks for Anti-Fraud Problems Solving
Maria A. Burnasheva, Elena A. Skorodumova · 2025
This study investigates the effectiveness of Kolmogorov-Arnold Networks (KAN) in detecting fraudulent transactions, comparing their performance with the CatBoost algorithm. The research utilizes a real-world dataset of over 284,000 credit card transactions, characterized by extreme class imbalance (only 0.0172% fraud cases). To mitigate this imbalance, the Undersampling technique was applied. Evaluation metrics including Precision, Recall, F1-score, and ROC-AUC were used to assess model performance. The findings reveal that KAN demonstrates superior precision for fraud detection (Class 1), making it particularly valuable for applications where false positives carry high costs. In contrast, CatBoost shows stronger performance in Recall and overall ROC-AUC, indicating better identification of actual fraud cases. While both models achieve high accuracy, KAN's balanced F1-score for the minority class suggests its potential for specialized anti-fraud tasks. The study concludes that model selection should be guided by specific operational requirements: KAN for precision-critical scenarios and CatBoost for maximizing fraud detection rates. These results contribute to the growing body of research on advanced machine learning techniques for financial security applications.