Abstract

The exponential growth of the Internet of Things (IoT) has led to theemergence of substantial security concerns, with IoT networks becoming theprimary target for cyberattacks. This study examines the potential ofKolmogorov-Arnold Networks (KANs) as an alternative to conventional machinelearning models for intrusion detection in IoT networks. The study demonstratesthat KANs, which employ learnable activation functions, outperform traditionalMLPs and achieve competitive accuracy compared to state-of-the-art models suchas Random Forest and XGBoost, while offering superior interpretability forintrusion detection in IoT networks.

 

Quick Read (beta)