Enhanced IoT Security through Balanced Machine Learning-Based Intrusion Detection Systems

Background/Objectives: In modern days, the Internet of Things (IoT) networks produce complex and severely imbalanced traffic because conventional intrusion detection systems are becoming less reliable. Method: This study has produced a CLIB-EBGM IDS, a calibrated ensemble intrusion detection framework that operates under extreme class imbalance while prioritising the reduction of false negatives and maintaining high predictive performance. The proposed technique integrates XGBoost, Random Forest, and Logistic Regression classifiers with leakage-aware preprocessing, SMOTE-based balancing, probability calibration, and an Energy-Balanced Cost Learning (EBCL) threshold optimisation strategy to speed up the attack detection. Here, the CALIB-EBGM IDS is run on the Bot-IoT benchmark dataset, and it is observed that it performs very efficiently. Findings: The result shows that the proposed CALIB-EBGM IDS achieves 99.96% accuracy and 99.95% precision. From the result comparison, it is observed that the proposed approach outperforms existing ensembles and traditional models, particularly when the dataset is highly imbalanced. Novelty: The combination of calibrated ensemble learning and cost-sensitive optimization enables a reliable IDS suitable for IoT gateways and edge devices. Keywords: Intrusion Detection System, Network Security, Machine Learning, Probability Calibration, Imbalanced Data, Attack Detection