PSO-Driven Feature Selection and Hybrid Ensemble for Network Anomaly Detection

As a system capable of monitoring and evaluating illegitimate network access, an intrusion detection (IDS) profoundly impacts information security research. Since machine learning techniques constitute the backbone IDS, it has been challenging to develop accurate mechanism. This study aims enhance performance IDS by using particle swarm optimization (PSO)-driven feature selection approach hybrid ensemble. Specifically, final subsets derived from different datasets, i.e., NSL-KDD, UNSW-NB15, CICIDS-2017, are trained ensemble, comprising two well-known ensemble learners, gradient boosting (GBM) bootstrap aggregation (bagging). Instead training GBM with individual learning, we train on subsample each dataset combine class prediction majority voting. Our proposed scheme led pivotal refinements over existing baselines, such as TSE-IDS, voting ensembles, weighted voting, other ensemble-based LightGBM.