Improved boosting-based machine learning algorithms for network intrusion detection in wireless sensor network

Abstrak

Intrusion detection is essential for protecting wireless sensor networks (WSNs) from evolving cyberattacks. This paper proposes an enhanced boosting-based framework that integrates generative adversarial networks (GANs) to address data imbalance, and Harris hawk optimization (HHO) for efficient feature selection. Six boosting algorithms, including adaptive boosting (AdaBoost), gradient boosting (GB), extreme gradient boosting (XGBoost), light gradient‑boosting machine (LightGBM), categorical boosting (CatBoost), and histogram-based GB, were evaluated to determine the most effective configuration. The proposed system achieves an accuracy of 99.18% with a detection time of 12.7 ms on a dataset for intrusion detection systems in WSN (WSN-DS dataset), significantly outperforming the existing boosting-based intrusion detection models. By combining data balancing and feature optimization, the framework enhances both accuracy and resource efficiency, providing a scalable and robust approach for real time threat detection in resource-constrained environments. The results confirm the potential of hybrid boosting methods coupled with advanced data generation and optimization strategies to strengthen the resilience of modern WSNs against emerging network attacks.