Enhanced Feature Extraction for IoT Network Intrusion Detection Using GNNs and KAN

The paper proposes SKGFusionKAN, an IoT-oriented intrusion-detection architecture built on GraphSAGE. It adds edge-oriented message passing, multi-scale selective-kernel attention, and gated feature fusion to adaptively model heterogeneous node and edge features. A Kolmogorov-Arnold Network (KAN) is then used for classification, targeting complex and low-frequency attacks. The authors state that the method consistently outperforms state-of-the-art approaches on binary and multiclass tasks across four NIDS benchmarks. However, the supplied abstract does not report dataset names, numerical results, ablations, or computational costs.