A Comprehensive Key Features Analysis and Recommendations based Cyber Intrusion Detection for Satellite-Terrestrial Networks

Abstract

The integration of Satellite-Terrestrial Networks (ISTN) necessitates advanced security measures, particularly Intrusion Detection Systems (IDSs). This study introduces hybrid sequential intrusion detection models for ISTNs, combining Deep Learning (DL) and Machine Learning (ML) techniques. The models employ both anomaly-based and signature-based detection to enhance accuracy, utilizing methods such as Extra Trees (ET), Decision Trees (DT), Random Forest (RF), XGBoost (XGB), and Gated Recurrent Units (GRU). These models are chosen for their superior performance and are used sequentially to improve IDSs effectiveness. RF-based Sequential Feature Selection (RF-SFS) is also utilized to reduce dataset dimensionality, which in turn decreases the computational costs for each model. Evaluations using UNSW-NB15 and STIN datasets-representing terrestrial and satellite traffic, respectively-demonstrate the models' superiority over traditional IDSs. The XGB-ET model achieved 99.99% accuracy in anomaly detection, while the XGB-GRU model attained 89% accuracy in signature-based detection on the UNSW-NB15 dataset. On the STIN dataset, the ET-DT-GRU model reached 96.47% accuracy in signature-based detection. Additionally, RF-SFS reduced execution times, with training and testing speedups up to 2.8x.