As cyber threats continue to grow in complexity and sophistication, the need for advanced network and sensor security solutions has never been more urgent. Traditional intrusion detection methods struggle to keep pace with the sheer volume of network traffic and the evolving nature of attacks. In this paper, we propose a novel machine learning-driven Intrusion Detection System (IDS) that improves intrusion detection through a comprehensive analysis of multidimensional data. Transcending traditional feature extraction methods, the system introduces geospatial context features and self-supervised semantic features that provide rich contextual information for enhanced threat identification. The system’s performance is validated on a carefully curated dataset from China Mobile, containing over 100K records, achieving an impressive 98.5% accuracy rate in detecting intrusions. The results highlight the effectiveness of ensemble learning methods and underscore the system’s potential for real-world deployment, offering a significant advancement in the development of intelligent cybersecurity tools that can adapt to the ever-changing landscape of cyber threats. Furthermore, the proposed framework is extensible to IoT and wireless sensor networks (WSN), where resource constraints and new attack surfaces demand lightweight yet semantically enriched IDS solutions.