Year: 2026
Type: conference-paper
Abstract: Wi-Fi networks have emerged as a critical communication medium for modern systems, making their security a growing concern. Despite widespread adoption of WPA3, transition-mode deployments continue to expose networks to downgrade attacks that revive KRACK vulnerabilities from the WPA2 era. This paper proposes an interpretable hybrid Convolutional Neural Network–Multilayer Perceptron (CNN–MLP) framework for detecting Key Reinstallation Attacks (KRACK) in WPA2/WPA3-compatible IEEE 802.11 Wi-Fi networks. Leveraging the AWID3 dataset, the model extracts discriminative features from packet-level data using CNN blocks and performs binary classification with MLP, achieving high performance metrics including 99.96% accuracy, a 99.45% F1-score, and a near-perfect AUC score. Interpretability is enhanced through techniques like integrated gradients and permutation importance, highlighting key features such as radio frequency and channel metadata. The lightweight design (0.112 MB model size) supports real-time deployment, addressing vulnerabilities in transition-mode networks.
Source: E3S Web of Conferences
Institutions: University of Uyo, Global Banking School
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7203794792