A novel gallop operator for horse herd optimization algorithm: application to feature selection of safety defect prediction in UAV

With the growing integration of Autonomous Vehicles (AVs) in military and civilian environments, ensuring robust security mechanisms, particularly through Intrusion Detection Systems (IDS), has become critical. However, the high dimensionality of UAV telemetry data often includes redundant or irrelevant features, which negatively impact the performance of machine learning algorithms. To address this, we propose an enhanced Binary Horse Herd Optimization Algorithm (BHOA) by incorporating a Gallop operator inspired by natural horse behavior. This operator improves the algorithm’s exploratory ability and avoids early convergence without increasing the computational burden. The enhanced BHOA is employed for feature selection, aiming to minimize redundancy and maximize relevance to the classification task. When coupled with the K-Nearest Neighbors (KNN) classifier for detecting spoofing and jamming attacks, the proposed method achieves an average of 8.25% improvement in classification accuracy compared to baseline feature selection algorithms such as minimum-Redundancy-Maximum-Relevance (mRMR) and ReliefF. Despite not converging faster in terms of iteration count, the Gallop-enhanced BHOA reaches higher-quality solutions in the same computational time. This model has potential real-world applications in UAV-based IDS systems, where efficient, lightweight, and accurate detection is essential for deployment under operational constraints. Future work will extend this framework to real telemetry data and explore domain adaptation to bridge the gap between synthetic training environments and real-world scenarios.