STATISTICAL AND MACHINE LEARNING BASED ATTACK DETECTION ON WIRELESS IoT NETWORKS

Abstract

The extensive use of Internet of Things technology over wide range of application domains brings with itself various security concerns. Most studies focus on complex and active approach with communication overhead among the deployed nodes, thus not capturing the key essence of an ideal defense system, mainly accuracy, speed, and low intrusiveness. In addition, the reliability of defense system’s design over signature and specification-based detection is infeasible considering the staunch heterogeneity of connected IoT devices. This thesis presents a set of security attacks and defense mechanisms leveraging the recent advances on attack detection methodologies in statistical and machine learning based detection. In this thesis, a physical layer-based defense system is designed focusing on 1) a novel multi-sequential testing approach to simultaneously detect illegitimate access and collision-based attack 2) a machine learning-based approach to detect smart Denial-of-Service and data manipulation attack, and 3) an improved one-class classification for anomaly detection.