A lightweight and robust approach to IoT intrusion detection based on ensemble deep learning

The inherent vulnerabilities and distributed nature of IoT systems necessitate IDS that balance accuracy with computational efficiency. This study proposes a lightweight, distributed IDS leveraging ensemble deep learning to address IoT security challenges. By integrating multiple lightweight models (DNN, CNN, GRU) via various ensemble strategies, the framework enhances detection capabilities while minimizing resource overhead. Evaluated on the CICIDS2017 and Edge-IIoT datasets, the ensemble approach outperforms single-model solutions, specifically stacking and boosting in rare threat detection. This work demonstrates the viability of decentralized, resource-aware IDS deployments, offering a scalable solution for securing dynamic IoT ecosystems amidst evolving cyber threats.