Fraud detection using Kolmogorov-Arnold Network (KAN)
In this study, we explored the Kolmogorov–Arnold Network for financial fraud detection. This approach was chosen for its interpretability, in contrast to traditional neural networks.
- Kaggle - Fraud detection using KAN
- Co-authors Nolack Tapsir Gislain Zeutouo, Evgeniy Yurievich Kostyuchenko
Context
Financial fraud represents a growing challenge for financial institutions and e-commerce platforms, requiring increasingly sophisticated detection methods. While traditional machine learning models have proven effective, they often struggle with complex fraud patterns.
Methodology
We used a synthetic dataset generated by the PaySim simulator, obtained from Kaggle. The data were preprocessed through downsampling to address class imbalance and by encoding categorical variables. Based on the processed data, we built a simple Kolgomorov-Arnold network architecture composed of two layers [5, 2].
Result
The resulting model achieved an F1-score of 97% for fraud detection, demonstrating its effectiveness in identifying fraudulent transactions.