Quantum Classical Hybrid Intrusion Detection For Open-Set IoT/IIoT Security

Abstract

The rapid evolution of cyber threats renders conventional intrusion detection systems increasingly ineffective against unknown attacks in IoT and IIoT networks. Although classical machine learning improves detection accuracy, these approaches struggle with open-set conditions, class imbalance, and resource limitations. In this thesis, we introduce QCH-IDS, a novel Quantum-Classical Hybrid Intrusion Detection System that combines parameterized quantum circuits (PQCs) with classical feature extraction and residual fusion, which enable robust hybrid representation learning under realistic deployment constraints. We design multiple configurations of a lightweight 4-qubit multi-head PQC architecture and evaluate it using a leakage-free SEEN/UNSEEN protocol across four different sets on the Edge-IIoTset and CICIoT2023 datasets. The fully hybrid QCH-IDS achieves near-perfect binary detection on Edge-IIoTset (99.99% in SEEN, and more than 95% in UNSEEN) and strong performance on CICIoT2023 (greater than 89%), outperforming expectations through feature reduction (59 features reduced to 46, and 46 features reduced to 39). The multiclass accuracy for SEEN (known attacks) exceeds 99% in Edge-IIoTset, and reaches 77% in CICIoT2023, showing competitive results for UNSEEN (unknown attacks) despite open-set challenges. We extended our evaluation to include precision, recall, F1-score, and distribution effects, where QCH-IDS consistently outperformed baseline expectations. Finally, we designed our model to be compatible with Noisy Intermediate-Scale Quantum (NISQ) by implementing it with 4 qubits and a shallow circuit depth. This ensures stability against quantum noise while maintaining high fidelity in near-term quantum hardware. This thesis demonstrates that hybrid models combining quantum and classical techniques provide significant advantages for detecting security threats in open-set IoT and IIoT traffic. It establishes clear and reproducible evaluation standards and paves the way for the development of practical intrusion detection systems that utilize quantum technology.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering