Enhancing IoT Security through Machine Learning-Based Intrusion Detection Systems

Objectives: This study targets the pressing limitations of traditional Intrusion Detection Systems (IDS) in IoT environments, notably the challenges posed by high-dimensional data, fluctuating detection accuracy, and elevated false alarm rates. Method: This study proposes a hybrid intrusion detection model that combines Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), and Deep Convolutional Neural Networks (DCNN). PSO and ABC are utilized for optimized feature selection, while DCNN performs hierarchical anomaly classification. The model is trained and evaluated on the UNSW-NB15 dataset, a widely accepted IoT intrusion benchmark. Findings: The proposed PSO-ABC-DCNN model achieves an accuracy of 99.41%, significantly outperforming existing models such as CNN (92.8%) and LSTM (98.2%). Novelty: The novelty lies in the integrated use of swarm intelligence techniques (PSO and ABC) with deep learning to create a lightweight, high-performance IDS. This synergistic approach enhances feature optimization, model generalization, and scalability in complex IoT environments. Keywords: IoT, Intrusion Detection, Machine Learning, Security, Cybersecurity, Supervised Learning, Unsupervised Learning, Deep Learning, Bot-IoT, UNSW-NB 15