SecIDS-CNN-WF: A trust-aware edge-efficient CNN for real-time Wi-Fi intrusion detection

Abstract

Wireless Fidelity (Wi-Fi) infrastructures continue to face increasingly dynamic and stealthy intrusion attempts, creating a strong need for efficient intrusion detection systems that operate reliably under strict edge–hardware constraints. This work introduces Secure Intrusion Detection System-CNN-Wi-Fi (SecIDS-CNN-WF), an adaptive and trust-aware intrusion detection framework that integrates semantic Hex–Word2Vec–KMeans–SMOTE (HWKS) preprocessing, an attention enhanced CNN backbone, and a zero-delay, multi-metric trust mechanism for real-time analysis of IEEE 802.11 traffic. The HWKS pipeline transforms raw traces into a structured representation of 155 numerical descriptors and 32 latent embeddings, improving class separability and stabilizing model optimization. Experimental results on the full five-class Aegean Wi-Fi Intrusion Dataset (AWID) show that SecIDS-CNN-WF achieves a macro-F1 of 99.97% and an overall accuracy of 99.96%, while sustaining approximately 0.20 ms per-sample inference latency and a compact 0.15 MB TensorFlow Lite footprint suitable for Raspberry Pi–class devices. The trust-scoring module increases alert reliability by filtering low-confidence outputs without adding temporal delay, and Integrated Gradients provide transparent feature level attribution over HWKS dimensions. Collectively, these results demonstrate that combining semantic featureization, attention based modeling, and trust-aware scoring enables real-time, interpretable, and resource efficient Wi-Fi intrusion detection on constrained edge platforms.

Peer Reviewed

Postprint (published version)