A Deep Learning Paradigm for Intrusion Detection in Unmanned Aerial Vehicle Networks Using Extended LSTM
Unknown authors
Abstract
Intrusion Detection Systems for Unmanned Aerial Vehicles (UAVs) networks face challenges with evolving cyber threats and network dynamics.Traditional machine learning and shallow neural networks suffer from low detection accuracy, high false positives, and an inability to capture temporal intrusion patterns in dynamic UAV environments with intermittent connectivity and limited computational resources.We introduce an Extended Long Short-Term Memory (XLSTM) framework, a novel deep learning-based IDS approach that integrates optimized memory cells, dynamic dropout regulation, and advanced recursive feature elimination (RFE) to address the challenges of UAV network intrusion detection.Unlike conventional LSTM and CNN-based IDS models, our method significantly enhances long-range dependency learning, reduces overfitting, and improves feature representation through adaptive feature selection.These enhancements enable robust anomaly detection with improved adaptability, computational efficiency, and resilience to adversarial attacks.Empirical evaluation demonstrates our model's superior performance with 99.87% accuracy, 99.91% precision, 99.65% recall, and 99.28% F1 score, validated using the CICIDS-2017 dataset.This dataset contains real-world network traffic simulations, including a diverse range of attack types such as Distributed Denial of Service (DDoS), brute-force attacks, botnet infiltration, and network scanning, alongside benign traffic instances.These results confirm XLSTM's capability to detect complex intrusion patterns with high accuracy while maintaining robustness against adversarial variations in UAV networks.