QUANTUM MACHINE LEARNING FOR REAL-TIME ANOMALY DETECTION IN IOT NETWORKS

Abstract

IoT is being rapidly adopted in multiple critical industries. Not only does it enable real-time monitoring and analysis of processes, but it also allows enormous optimizations and automation. From manufacturing, logistics and healthcare to smart homes and cities, the applications of IoT are expanding rapidly, leading to more connectivity and digital communication of smart devices over the networks. Although IoT systems have significantly increased productivity, they have also raised serious concerns about the security of such systems, leading to a growing need for fast and accurate intrusion detection systems that can flag abnormal behaviors in these systems, in real-time. Anomaly detection in real-time is a crucial shield against data breaches, malicious attacks, and abnormal system failures. In the present day, where IoT ecosystems are becoming highly complex, classical anomaly detection models often struggle with scalability and accuracy. There is both a need and an opportunity to adopt and deploy more sophisticated and accurate anomaly detection systems. Anomaly detection systems have traditionally relied on signatures and basic statistical approaches, which have proven to be ineffective. Then, machine learning-based approaches emerged to address these inefficiencies, significantly improving the accuracy of such systems. However, these approaches still face constraints in terms of resources and data. Quantum Machine Learning has recently emerged as a potential solution to overcome these limitations and inefficiencies. QML combines traditional machine and quantum computing, leveraging the power of quantum systems, and has evolved as a promising approach for detecting anomalies in IoT and networks. Models based on quantum-classical hybrid learning have demonstrated the ability to perform better than classical machine learning models. In this thesis, we propose a Quantum-Classical Hybrid Learning-based model for real-time anomaly detection in IoT networks. We use two IoT/network public datasets to train and test the model and validate the results through multiple cross-fold validation experiments. The results of experiments not only show promising results in detecting anomalies, but have also outperformed previous quantum machine models and classical machine learning models that were trained on the same datasets, potentially highlighting the potential of quantum-classical architecture in improving the security of IoT devices.

Date of Award	2025
Original language	American English
Awarding Institution	HBKU College of Science and Engineering