Metaverse-aware intrusion detection system for digital twin-based enterprise systems using entropy-aware sparse attention transformer

Enterprise systems in Industry 4.0 increasingly rely on digital twins and metaverse integration for real-time monitoring, control, and decision-making. However, these systems remain vulnerable to evolving cyber threats, and traditional attention-based models used in intrusion detection suffer from high computational complexity (O(n2)). To address this, we propose a sparse attention-based transformer model that used entropy for the calculation of attention. The model combines categorical embeddings with continuous features and applies entropy-guided sparse attention to reduce complexity and enhance feature relevance. Using the CIC-DDoS2019 dataset, the model achieves 95.24% precision and high AUC scores (0.99) while using significantly fewer trainable parameters.