AI-Based Intrusion Detection in IoT Networks Using Lightweight Deep Learning Models

The exponential growth of the Internet of Things (IoT) ecosystem has revolutionised automation and connectivity across diverse domains, but it has also amplified cybersecurity vulnerabilities due to the heterogeneous, large-scale, and resource-constrained nature of IoT devices. Traditional intrusion detection systems (IDS) struggle to achieve scalability, low latency, and real-time adaptability in such dynamic environments. This paper proposes a lightweight deep learning-based intrusion detection framework tailored for IoT networks, emphasising computational efficiency, high detection accuracy, and Model interpretability. The proposed architecture integrates optimised convolutional and recurrent modules with attention mechanisms for effective spatial–temporal feature extraction while maintaining a minimal parameter count, making it suitable for deployment on edge and embedded devices. Unlike conventional heavy models such as OSEN-IoT and CST-AFNet, the proposed framework balances accuracy and efficiency by leveraging reduced-parameter neural blocks inspired by Kolmogorov–Arnold Networks (TFKAN) and TinyML optimisation strategies. Extensive evaluation on benchmark datasets, including BoT-IoT, ToN-IoT, and CICIoT2023, demonstrates that the proposed model achieves detection accuracy exceeding 99%, with a false positive rate below 0.2%, outperforming existing approaches such as CAEAID, Ex3WNN, and FedMSE while reducing computational overhead by more than 65%. This research contributes to the development of scalable, interpretable, and lightweight intrusion detection systems capable of securing large-scale IoT deployments in real time.