AegisGuard: A Progressive Quantum‐Enhanced Hybrid Intrusion Detection System for Industrial Internet of Things Security

Smart-grid Industrial Internet of Things (IIoT) deployments face fast-evolving cyber-threats, extreme class imbalance, and tight real-time constraints that degrade the reli-ability of conventional intrusion detection systems (IDS). Existing IDS frameworks of-ten struggle with generalizability across heterogeneous IIoT datasets and fail to simul-taneously balance accuracy, efficiency, and low false alarms. This paper presents a system named AegisGuard, a sophisticated hybrid intrusion detection system that combines a four-stage sampling pipeline with a calibration ensemble learner specifi-cally designed for IIoT traffic. The pipeline systematically combines SMOTE, SMOTE-ENN, ADASYN, and strategic undersampling to counter severe imbalance (≈99.8% benign), while a quantum-inspired feature selection scheme fuses F-test, mu-tual information, and random-forest importance with trust-aware weighting to retain 25 of 46 features. The final classifier ensembles Random Forest, Extra Trees, LightGBM, XGBoost, and Catboost with Optuna-guided tuning and post-hoc probability calibra-tion, optimized under a composite objective that jointly minimizes false alarms. On CIC IoT 2023, AegisGuard lifts accuracy from a 5-model baseline of 89.6% to 99.6%, cutting the false alarm rate (FAR) to 0.31%. Cross-dataset evaluation demonstrates robustness: TON-IoT achieves 98.3% accuracy (FAR 0.4%), UNSW-NB15 98.4% (FAR 1.1%), and Bot-IoT 99.4% (FAR 0.8%). Dimensionality reduction yields a 54% feature cut and a 65% memory reduction (to 2.3 GB), with sub-second inference (0.42 s/sample) suitable for operational monitoring.