Optimizing Energy-Efficient Machine Learning Algorithms for Real-Time Attack Detection in IoT Devices

Abstract

Energy efficiency is a critical challenge in the expanding domain of In-ternet of Things (IoT) networks, where resource-constrained devices must operate securely under strict energy limitations. This study explores the application of Levy-Based Moth-Flame Optimization (LB-MFO) to enhance intrusion detection in IoT systems. Using the CICIDS 2017 dataset, LB-MFO was evaluated against standard machine learning models, including Logistic Regression, Decision Trees, Random Forest, and Support Vector Machines. To further optimize energy usage, techniques such as pruning, quantization, and model compression were employed. The results demonstrate that LB-MFO consistently outperformed other models, achieving 96% accuracy with a performance score of 0.98 while consuming only 0.018–0.020 W with minimal latency of 14–15 ms under optimization. These findings highlight LB-MFO’s potential to deliver accurate and energy-efficient intrusion detection, making it ideal for real-time IoT applications.