Quantum Machine Learning for Network Traffic Classification | Dr. Fahad Ahmad | ICQIA 2026

The interesting question is not whether quantum models beat classical ones. It is what they achieve on a fraction of the resources. In this ICQIA 2026 session, Dr. Fahad Ahmad, Senior Lecturer at the School of Computing, Mathematics and Physics, University of Portsmouth, United Kingdom, and visiting researcher in the Netherlands, presents work comparing classical and quantum approaches to intrusion detection in network traffic. Modern attacks are fast, adaptive and hidden inside massive heterogeneous traffic, and the environments that most need detection are often the ones with the least available compute. That framing sets his research question: not whether quantum machine learning outperforms classical methods outright, but whether it can identify intrusion patterns under real resource constraints. The study uses the CIC IoT 2023 benchmark dataset, classifying traffic into reconnaissance, spoofing and other attacks. Classical models, a convolutional neural network, a residual model and an attention model, use all 39 features and around 74,000 instances. The quantum models compress those 39 features to eight via principal component analysis to fit an 8-qubit budget, and train on roughly 2,000 instances. To compensate for the qubit limit, he uses a data re-uploading model, encoding the input features multiple times through the circuit rather than once, so the model can learn the structure of the data without adding qubits. Data re-uploading outperforms the baseline quantum neural network, 68.40 against 67, while the best classical model reaches 84.10. His conclusion: numerically the classical models are ahead, but the resource gap is far larger than the performance gap. Given comparable features, samples and compute, and running on real hardware rather than simulators, the quantum results look promising. In this session: 0:00 Introduction 0:55 Agenda 2:39 Why this problem matters 4:38 The research question 5:28 Dataset, features and attack categories 6:18 Classical versus quantum pipeline 8:29 The full experimental pipeline 9:18 The data re-uploading model 10:26 Why repeated encoding works under qubit limits 11:56 Results 14:38 Interpreting the resource gap 15:16 Limitations and next steps on real hardware 16:08 Key takeaway ICQIA 2026, the International Conference on Quantum Intelligence and AI Applications, is organised by Entipivi Digital Advisory. Conference: https://entipivi.com/icqia2026.html Advisory and training: https://entipivi.com LinkedIn:   / entipivi   #Quantummachinelearning #networksecurity #entipivi #ICQIA2026