An Energy-Efficient Framework for Real-Time Anomaly Detection and Threat Mitigation in IoT Traffic Streams

Abstract

The integration of IoT devices in smart city traffic systems enhances real-time monitoring but increases vulnerability to cyber threats such as data breaches and zero-day attacks. Traditional intrusion detection systems (IDS) are often energy-intensive and ineffective against evolving threats, prompting the need for adaptive, energy-efficient security frameworks. This paper presents an energy-efficient framework for real-time anomaly detection and threat mitigation in IoT traffic streams. It combines lightweight machine learning models (RF, SVM, LR, XGBoost) with deep learning (CNN Autoencoder) for accurate, low-overhead anomaly detection. Simulated Annealing is employed for hyperparameter optimization, reducing computational load. The framework also integrates Blockchain-based identity verification and a lightweight Zero Trust Architecture (ZTA) to secure access control. Experiments using the IoT-23 dataset demonstrate superior accuracy, reduced false positives, and lower energy usage compared to traditional methods, making the proposed solution suitable for secure, scalable, and energy-aware smart city deployments.