An Efficient Anomaly Detection System in IoT Edge Using Chi–Square and Improved Particle Swarm Optimization Feature Selection with Ensemble Classifiers

Anomaly detection using machine learning (ML) has become a critical research focus point in the modern digital era. Although recent ML-based models demonstrate strong detection capabilities, their performance is often limited by the vast volume and high dimensionality of data, leading to reduced accuracy, lower detection rates, and increased learning complexity. To address these challenges, this study proposes a novel ensemble learning framework that combines random forest (RF), extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), category boosting (CatBoost), and a light gradient boosting machine (LGBM) with a hybrid feature selection strategy. An improved particle swarm optimization (IPSO) algorithm enhanced with elimination and opposition-based learning is developed and integrated with the chi-square method (chi-IPSO) to optimize feature selection. Additionally, the synthetic minority oversampling technique (SMOTE) is employed to handle class imbalance in the datasets. The proposed framework is evaluated using two benchmark datasets, UNSW-NB15 and CICIDS 2017. Experimental results reveal that the RF algorithm with chi-IPSO achieves superior performance, attaining an accuracy of 94.58% on UNSW-NB15 and 99.70% on CICIDS 2017. These results illustrate the effectiveness of the proposed model, showing notable betterment over existing state-of-the-art anomaly detection approaches.