Answers are grounded in this paper and cite their sources.
Network intrusion detection systems (NIDS) are essential for identifying malicious traffic, but many deep learning approaches achieve high detection performance at the cost of large models and high computational complexity. This paper proposes a compact quantum convolutional neural network (QCNN) framework for binary and multiclass intrusion detection using CSE-CIC-IDS2018 and UNSW-NB15. Eight traffic features selected by training-set mutual information are encoded using RY rotations. The binary QCNN has 11 trainable parameters and a circuit depth of 16. For multiclass detection, a fixed circuit produces 16 joint measurement probabilities for Random Forest classification (QCNN-RF). Across three runs on fixed partitions, mean binary test accuracy was 98.92 ± 0.29% on CSE-CIC-IDS2018 and 99.17 ± 0.29% on UNSW-NB15. The corresponding attack-class F1-scores were 98.92 ± 0.29% and 99.17 ± 0.28%. Multiclass accuracy and macro-F1 were 92.44% and 91.98% on CSE-CIC-IDS2018 and 94.30% and 94.30% on UNSW-NB15. On CSE-CIC-IDS2018, the fixed transformation increased mean macro-F1 by 4.08 percentage points relative to a Random Forest using the same eight input features directly. Circuit sizes and simulator runtimes are reported separately to distinguish architectural compactness from computational cost.
No takes yet. Share an insight, caveat, or question.
Kaissar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: