Cross-Layer Analysis of Machine Learning Models for Secureand Energy-Efficient IoT Networks

• DOI: 10.20944/preprints202504.1349.v1
• Publication Date: 2025-04-18T03:50:12Z
• Authors: Rashid Mustafa; Nurul I. Sarkar; Mahsa Mohaghegh; Shahbaz Pervez; Ovesh Vohra
• Keywords: Application layer

The widespread use of Internet Things (IoT) raises security and energy efficiency 1 concerns, particularly for low-resource devices. In this paper, we analyse a cross-layer IoT architecture 2 using machine learning models lightweight cryptography. We focus on analysing vulnerabilities 3 suggest energy-efficient solutions. Our proposed solution is based role-based access 4 control ensuring secure authentication in large-scale deployments blocks undesired 5 attempts. By combining convolutional neural networks, rule-based systems, hybrid artificial 6 intelligence, the improves accuracy anomaly identification 7 while lowering false positives. system performance evaluated by simulations as well testbeds 8 to accomplish attack mitigation. Results show that reduces positives 9 28–32% provide improved preventing 95% unwanted access. found up 30% 10 power reduction Speck encryption (8Hz ContikiMAC duty cycle) than 11 traditional AES encryption. For data injection, sinkhole jamming attacks, system’s 12 resilience confirmed Cooja/Contiki simulations, which maintain packet delivery rate. 13 from networks our approach efficiently 14 practical scenarios such smart schools.