PEARL: A Reinforcement Learning-Based Attack Analysis Approach for Security and Robustness Assessment of Smart Home Systems
Book Chapter
Haque, NI, Rahman, MA, Uluagac, S et al. (2026). PEARL: A Reinforcement Learning-Based Attack Analysis Approach for Security and Robustness Assessment of Smart Home Systems
. 688 LNICST 127-150. 10.1007/978-3-032-23450-6_7
Haque, NI, Rahman, MA, Uluagac, S et al. (2026). PEARL: A Reinforcement Learning-Based Attack Analysis Approach for Security and Robustness Assessment of Smart Home Systems
. 688 LNICST 127-150. 10.1007/978-3-032-23450-6_7
False data injection (FDI) in sensor measurements within smart homes can result in increased energy expenditures and compromise the health and safety of occupants. Machine learning (ML)-based anomaly detection models (ADMs) are extensively integrated into smart home control systems to enhance overall security and counter FDI attack impact. However, an adversary possessing sufficient knowledge of the system and computational resources can exploit vulnerabilities within control models and ADMs to orchestrate stealthy FDI attacks. Hence, modern ADMs are assessed against stealthy FDI attacks instead of random attacks for realistic security and robustness analysis. Nevertheless, identifying the stealthy attack vectors from ML-based non-linear ADMs is yet to be contemplated. We propose a novel reinforcement learning-based framework, PEARL, to identify stealthy attack vectors from a non-linear ADM-assisted smart home control system, which overcomes the limitations of state-of-the-art (SOTA) analytics. We validate our framework using the real-world Activity Recognition with Ambient Sensing (ARAS) dataset.
