Optimized AI-driven Malware detection and network attack identification using a lightweight gradient ensemble in IoT edge

The expanding usage of IoT devices requires resource-efficient cybersecurity solutions because they operate with limited capacity. This work presents LGWEM as a framework for detecting network attacks in IoT environments which focuses on lightweight performance. The model enhances classification accuracy and decreases computation requirements through its voting mechanism of feature selection using random forest alongside extra tree classifiers and K-best selection. The integrated method of learning outperforms traditional models when XGBoost and Bayesian optimisation are used to adjust variables. The proposed evaluation of the dataset using UNSW-NB15 showed that the approach achieved superior performance results specifically through utilization of the selected top ten features. The study presents two optimization techniques which include gradient-based one-side sampling (GOSS) and exclusive feature bundling (EFB) to maximize training efficiency. Secure IoT operations become achievable through edge attack detection in real-time that decreases the need for cloud computing resources. The study delivers a flexible and resource-efficient cybersecurity system intended for IoT networks which resolves major intrusion detection issues while using minimal system resources. Data shows LWGEM reaches 99% detection accuracy through decreased computational expenses together with a decreased occurrence of false positives.