Fraud Detection in Supply Chain Order Management via Kolmogorov–Arnold Networks

This study proposes a novel approach that utilizes Kolmogorov-Arnold Networks (KAN) for detecting fraud in trade orders within supply chains. The increasing prevalence of order fraud in supply chains poses a significant threat to the economic interests and reputations of organizations. Traditional fraud detection methods have limitations when addressing high-dimensional nonlinear data; however, the KAN model, with its superior nonlinear fitting capabilities and ability to learn complex data patterns, has emerged as a powerful tool for tackling this issue. In this study, we employ a high-dimensional dataset comprising a substantial number of actual trade orders and construct a KAN model for order fraud detection through systematic data cleaning and feature extraction. The experimental results demonstrate that the KAN model significantly outperforms traditional machine learning methods and other neural network models in identifying fraudulent orders, excelling in several evaluation metrics, including accuracy and F1 score. Furthermore, the KAN model exhibits enhanced robustness in addressing the data imbalance problem and effectively reduces the false alarm rate. This study not only validates the application potential of the KAN model in supply chain risk management but also offers new ideas and methodologies to further advance fraud detection technology.