A Quantum Inspired Horse Herding Optimizer and graph transformer for efficient Network Intrusion Detection Systems

Intrusion Detection Systems (IDS) play a major role in identifying and protecting against abnormal network actions. Specifically, anomalous or malicious behaviour was detected using network traffic patterns in the Network Intrusion Detection System (NIDS). However, the challenges in the NIDS are false alarm rates and achieving a high detection accuracy. To overcome these challenges, Quantum-Inspired Horse Herding Optimizer (QIHHO) with Graph Transformer Network is proposed in this study for effective NIDS. This approach helps to address the limitations in traditional metaheuristics, which are premature convergence and insufficient exploration. QIHHO enables subset selection and efficient global search features by utilizing quantum superposition, probabilistic encoding, and an adaptive rotation mechanism. This method enhances the discriminative performance by reducing the computational cost and feature redundancy. Additionally, the long-range temporal dependencies were captured by using the transformer decoder, where utilizing multi head self-attention. On the other hand, Graph Neural Network (GNN) encoder models are aided by spatial communication relationships with network entities. This combination facilitates precise identification of coordinated, low-frequency, multi-stage incursions. Extensive tests with three benchmark datasets (NSL-KDD, UNSW-NB15, and CICIDS2017) show that the suggested framework performs better. QIHHO-GTrNIDS achieves 98.42%, 99.03%, and 99.12% accuracy on these datasets, respectively, while maintaining an exceptionally low False Alarm Rate (FAR) of 0.0108, 0.0097, and 0.0091.