End-to-End Intrusion Detection in Data Security Using Attention-Based Large Language Model

The critical importance of cybersecurity escalates alongside the rapid progress in information technology, with data security intrusions posing severe threats to personal privacy and enterprise system integrity. Conventional intrusion detection systems, particularly challenged by complex, dynamic networks and diverse attack vectors, frequently exhibit insufficient detection accuracy coupled with elevated false alarm rates. Confronting these limitations, this paper introduces a deep learning-based data security intrusion detection system. This novel solution integrates the Mamba model for foundational feature extraction, establishing efficient data representations. Subsequently, the ECANet model refines feature selection via its attention mechanism, dynamically prioritizing the most critical features. The entire architecture undergoes end-to-end learning for holistic training and optimization, ensuring robust real-world applicability. Experimental validation confirms the system’s superior performance, consistently attaining a 5% higher detection accuracy across varied test datasets compared to traditional methods, thereby presenting an effective innovation for data security protection.