A Hybrid Optimization Framework for Enhancing IoT Security via AI-based Anomaly Detection

Abstract

With the rising integration of IoT devices in critical applications, ensuring real-time and robust security has become a major challenge due to limited computational resources and increasingly sophisticated cyber threats. This study introduces a hybrid optimization framework that leverages AI-based anomaly detection enhanced by metaheuristic optimization techniques. The system combines deep learning models—such as autoencoders and LSTM networks—for effective anomaly identification, with a hybrid tuning strategy using Genetic Algorithms (GA) and Ant Colony Optimization (ACO) to optimize model parameters, feature selection, and detection thresholds. The multi-objective optimization approach balances detection accuracy, computational efficiency, and false alarm reduction, making it suitable for diverse IoT environments including smart homes, healthcare, and industrial networks. Experimental evaluations on real-world IoT datasets reveal that the hybrid framework significantly outperforms standalone AI or optimization methods in threat detection reliability and energy efficiency. This research contributes a flexible, high-performance security architecture tailored for the next generation of secure, intelligent IoT systems.