Multi Layer Jamming Detection and Classification in 6G-Enabled Satellite-Air-Ground Integrated Networks Using Machine Learning

Future 6G networks aim to ensure ultra-reliable, high-throughput communication not only in dense urban environments but also across remote, rural, and disaster-prone regions. To support such ubiquitous connectivity, the Satellite-Air-Ground Integrated Network (SAGIN) has emerged as a promising multilayer architecture, integrating Low Earth Orbit (LEO) satellites, High Altitude Platform Stations (HAPS), Unmanned Aerial Vehicles (UAV), and terrestrial networks. However, the heterogeneous and dynamic nature of SAGIN significantly expands its attack surface, with multi layer jamming attacks posing a severe threat to communication reliability and network security. These jamming attacks can originate from space, air, or ground layers and may simultaneously target multiple communication domains, disrupting both data and control links. Despite their critical impact, research on the characteristics and correlations of multi layer jamming remains scarce, largely due to the lack of integrated platforms and publicly available datasets. Additionally, varying conditions across layers—such as mobility, power constraints, and spectrum usage—make real-time jamming classification extremely challenging. Traditional threshold-based approaches often result in high false alarm rates and poor generalization. To address these issues, we proposed a novel random forest based supervised machine learning classification model capable of real-time detection, source and target identification, and jamming type classification in SAGIN. To train the proposed random forest algorithm, we propose a novel synthetic dataset generation framework that models coordinated jamming scenarios under diverse SAGIN conditions. Moreover, this paper presents the first in-depth analysis of multi layer jamming characteristics and inter-layer correlations. Comprehensive evaluation results—comprising statistical analysis, correlation studies, confusion matrices, performance metrics, robustness assessment, and benchmarking against state-of-the-art machine learning models—validate the effectiveness, robustness, and generalization capability of the proposed framework.