Fraud Detection using Kolmogorov-Arnold Network

Abstract

With the rise of mobile payment services, new opportunities for illegal activities and fraud have emerged, prompting technology companies and financial institutions to prioritize rapid fraud detection. This research presents a financial fraud detection model based on the Kolmogorov-Arnold Network (KAN); a novel neural network technique developed by MIT researchers in 2024. By using the synthetic financial for fraud detection Paysim dataset from Kaggle, we achieved notable performance metrics: 97% precision, 97% recall, 97% F₁ score and 97% accuracy. These impressive results indicate that the KAN model can effectively identify fraudulent transactions with a high degree of reliability. For banks, this means enhanced security measures that can significantly reduce financial losses due to fraud. By implementing such advanced detection systems, financial institutions can not only protect their assets but also build greater trust with their clients.