Akademik Veri Yönetim Sistemi
Journal of Supercomputing, cilt.82, sa.2, 2026 (SCI-Expanded, Scopus)
This paper addresses the growing cybersecurity challenges in Agricultural Internet of Things (AG-IoT) environments, where large-scale sensor networks generate high-dimensional and complex traffic data that traditional machine learning (ML) models struggle to analyze effectively. To improve intrusion detection performance under these conditions, we propose a hybrid quantum–classical neural network model designed to capture complex patterns in AG-IoT traffic. High-dimensionality is first reduced using a binary Starfish Optimization Algorithm (SFOA) for feature selection, inspired by starfish foraging and regeneration behaviors. SFOA was chosen due to its superior convergence performance benchmarked against 100 modern metaheuristic algorithms, making it highly suitable for large AG-IoT datasets. A secondary mutual-information-based reduction further selects the most informative features for quantum processing. These features are then encoded into quantum circuits and evaluated using three architectures: a fully quantum model, a hybrid model combining quantum feature transformation with a classical classifier, and an enhanced hybrid model incorporating additional classical layers. All quantum experiments were executed on noise-free quantum simulators rather than physical hardware. Experiments conducted on the publicly available Farm-Flow dataset, which includes 1.3 million AG-IoT traffic flows across eight attack categories, demonstrate that the proposed hybrid approach achieves 90.10% accuracy in binary classification and 84.60% accuracy in multiclass intrusion detection. These findings suggest that the model provides performance comparable to conventional ML methods reported in the literature. Ultimately, this study highlights that quantum machine learning (QML), when paired with optimized feature selection, offers a promising and effective direction for securing AG-IoT systems against evolving cyber threats.