Secure IoT sensor networks through advanced anomaly detection with Kolmogorov-Arnold Networks (KANs)

In the field of the Internet of Things (IoT), the security and reliability of sensor networks is rather lacking, open to Denial of Service (DOS), scanning, malicious control, malicious operation, spying, and data probing. This paper explores the use of Kolmogorov-Arnold Networks (KANs) for advanced anomaly and attack detection in IoT sensor networks. For real-time application, the implementation makes use of Gaussian Radial Basis Function (RBF) along side with Reflectional Switch Activation Function (RSWAF). The RBFs allows the network to capture local non-linear relationships, improving the performance of model, both in terms of accuracy and computational efficiency. The RSWAF provides a computationally efficient activation mechanism that facilitates faster learning and inference. Our experiments demonstrate that the faster-KAN implementation significantly reduces training and inference times while maintaining high accuracy and robustness in detecting anomalies and attacks, achieving an accuracy of 99.38% for test dataset. Other metrics, such as F1-score, precision, recall, and confusion matrix are used to estimate the effectiveness of the model and compare with other models.