Adversarial Attacks Resilient Machine Learning-Enabled Access Control Scheme for Consumer Electronics in IoT-Based Applications

Consumer Electronics (CE) have become integral to the evolving Internet of Things (IoT) ecosystem, providing enhanced security, interoperability, and intelligence through machine learning (ML) to improve IoT functionality and user experiences. However, as CE devices manage sensitive data, security vulnerabilities emerge, particularly due to the use of wireless communication channels, which are often insecure. The lack of security in ML training and testing on CE-related datasets heightens risks such as data poisoning and adversarial attacks. In this article, a resilient machine learning-based access control scheme (RMACE-IoT) is designed to address these critical security challenges in IoT applications. RMACE-IoT not only strengthens the security of data exchanges but also defends against adversarial attacks that could undermine ML model predictions. The proposed RMACE-IoT is validated through a real-time testbed experiment, showcasing its effectiveness in enhancing resilience and mitigating adversarial threats. The proposed scheme (Case 1) achieves communication cost reductions of up to 40.00%, 60.42%, 61.49%, 65.87%, and 81.73% when compared with the schemes of Luo et al., Ali and Pal, Wang et al., Ever, and Zhang et al., respectively. In terms of computational cost, the proposed scheme demonstrates improvements of 99.90%, 52.76%, 52.29%, 99.95%, and 45.70% over the corresponding schemes. These results demonstrate the practical applicability and advantages of RMACE-IoT in real-time IoT scenarios. This work significantly advances the field by addressing pressing security concerns in CE systems, paving the way for safer, more reliable, and intelligent consumer devices.