A Hybrid CNN–Transformer Gated Fusion Intrusion Detection System for Edge-Enabled Smart Homes

DOI: https://doie.org/10.10399/JBSE.2025328313

Khaja Moinuddin, Prabhavathi S

Keywords:

IOT Security, Intrusion detection, Transformer, Gated fusion. Smart home

Abstract:

Smart homes rely on interconnected IoT devices, increasing vulnerability to sophisticated cyber attacks. Intrusion Detection Systems (IDS) are essential for safeguarding the IoT ecosystem. Although recent deep learning models have achieved high accuracy in intrusion detection, their substantial computational requirements hinder deployment on resource-constrained IoT devices. To address this challenge, we propose CTGF-IDS (CNN–Transformer Gated Fusion Intrusion Detection System), a lightweight yet powerful intrusion detection framework. CTGF-IDS integrates convolutional layers for efficient local feature extraction with Transformer-based attention mechanisms for modelling long-range dependencies, while a gated fusion module optimizes multi-scale feature integration. This architecture minimizes computational overhead and enhances on-device resource utilization without sacrificing accuracy. CTGF-IDS achieves a substantial reduction in floating-point operations, enabling real-time detection with improved precision, speed, and energy efficiency. Furthermore, knowledge distillation strengthens CTGF-IDS against shifts in traffic distribution, ensuring adaptability in dynamic IoT environments. We evaluate CTGF-IDS using the BoT-IoT and CIC-IDS2017 benchmark datasets, and results demonstrate superior performance compared with existing intrusion detection approaches. The model significantly reduces parameter count, FLOPs, and memory footprint while maintaining high detection accuracy across diverse network traffic patterns. By combining efficiency, robustness, and scalability, CTGF-IDS advances next-generation intelligent IoT security.

PDF