Machine Learning-Based Physical Layer Security for Detecting Active Eavesdropping Attacks

Abstract

This letter explores machine learning for enhancing physical layer security in a wireless system with an access point, legitimate users, and an active eavesdropper. During uplink training, the eavesdropper mimics pilot signals to compromise communication. We propose a framework to extract statistical features from wireless signals and build physical layer datasets. A one-class Support Vector Machine (OC-SVM) is used to detect such active eavesdropping attacks. Additionally, we introduce a twin-class SVM (TC-SVM) model to evaluate and compare detection performance. Simulation results demonstrate that our proposed approach with OC-SVM achieves a detection accuracy of 99.78%, performing favorably compared to the TC-SVM model and other prior methods.