Publication

Abstract :

The rise of edge computing for latency-sensitive applications highlights security concerns in resource-constrained IoT devices, necessitating efficient and lightweight IDS to protect data integrity and confidentiality while minimizing resource usage. However, many existing IDS solutions are not optimized for such environments, leading to inefficiencies in real-time threat detection. To address these challenges, this work develops a lightweight hybrid multimodal IDS for intrusion detection in edge-enabled IoT environments. By analyzing data from various sensors–such as temperature, humidity, pressure, and acceleration–the proposed IDS accurately identifies known and unknown attacks, unlike traditional unimodal approaches that often fail to capture threats from diverse sources. Our approach leverages an ensemble voting mechanism combining LGBM and XGBoost, considering their efficiency and scalability, ensuring a more robust decision-making process. Evaluation using the Edge-IIoT dataset shows that our proposed IDS approach outperforms other models, achieving a higher accuracy of 98.51% and 88.41% for binary and multiclass classification, respectively, with minimal CPU utilization. To enhance the performance and guarantee its lightweight nature, an efficient feature selection using variance threshold filter and Bayesian optimisation involving hyperparameter tuning are utilized, reducing power consumption and optimizing hardware usage. The IDS is deployed on a Raspberry Pi 4, demonstrating its suitability for resource-constrained edge devices with an average CPU utilization of 17-18%, significantly lower than standalone models. The model’s energy efficiency is validated using a B2901A Precision Source/Measure Unit (SMU), confirming reduced power consumption compared to existing approaches. Our research highlights the potential of adaptive IDS for real-time, scalable, and lightweight threat detection in edge-enabled IoT environments.