<![CDATA[Purpose – This study aims to develop a robust multiclass intrusion detection system (IDS) for Internet of Things (IoT) environments by optimizing Light Gradient Boosting Machine (LightGBM) using Particle Swarm Optimization (PSO), with a focus on improving performance under severe class imbalance. Design/methods/approach – A PSO-based hyperparameter tuning framework is applied to LightGBM, where Macro F1-score is used as the fitness function to ensure balanced class performance. The model is evaluated on the RT-IoT2022 dataset using a leakage-safe stratified 70:15:15 split. Performance is assessed using Accuracy, Macro Precision, Macro Recall, Macro F1-score, Weighted F1-score, and Matthews Correlation Coefficient (MCC). Experiments are repeated across 10 runs, and statistical significance is validated using the Wilcoxon signed-rank test. Findings - The proposed PSO-LightGBM model significantly outperforms the baseline LightGBM. It achieves 99.75% accuracy, 97.10% macro F1-score, 99.75% weighted F1-score, and 99.37% MCC, compared to 85.36%, 25.44%, 84.23%, and 63.67%, respectively, for the baseline. The model demonstrates substantial improvement in minority-class detection, reducing misclassification and preventing class collapse observed in the baseline. Research implications/limitations – The findings highlight the effectiveness of Macro-F1-guided optimization for imbalanced multiclass IoT intrusion detection. However, the evaluation is limited to a single dataset and centralized experimental setting, which may affect generalizability. Originality/value – This study contributes a leakage-safe, Macro-F1-driven PSO-LightGBM framework with comprehensive evaluation, including class-wise analysis, repeated runs, and statistical testing, providing strong evidence for balanced multiclass IoT intrusion detection.]]>
