The growing complexity of modern network infrastructures, driven by cloud computing, IoT, and 5G, poses significant challenges for traditional Intrusion Detection Systems (IDS), which often fail to detect novel and sophisticated cyber threats. This study proposes a Transformer-based IDS architecture, leveraging attention mechanisms to model temporal dependencies and contextual relationships in network traffic. Using benchmark datasets (NSL-KDD, CICIDS2017, UNSW-NB15), the proposed FT-Transformer demonstrates superior performance over conventional CNN, BiLSTM, and vanilla Transformer models, achieving an F1-score of 97.7% with strong generalization across diverse attack types. The research addresses critical issues, including class imbalance, real-time detection, and the integration of explainable AI techniques. Results confirm that Transformer architectures offer a scalable and robust solution for modern IDS frameworks, with potential for deployment in enterprise, cloud, and edge...
The growing complexity of modern network infrastructures, driven by cloud computing, IoT, and 5G, poses significant challenges for traditional Intrusion Detection Systems (IDS), which often fail to detect novel and sophisticated cyber threats. This study proposes a Transformer-based IDS architecture, leveraging attention mechanisms to model temporal dependencies and contextual relationships in network traffic. Using benchmark datasets (NSL-KDD, CICIDS2017, UNSW-NB15), the proposed FT-Transformer demonstrates superior performance over conventional CNN, BiLSTM, and vanilla Transformer models, achieving an F1-score of 97.7% with strong generalization across diverse attack types. The research addresses critical issues, including class imbalance, real-time detection, and the integration of explainable AI techniques. Results confirm that Transformer architectures offer a scalable and robust solution for modern IDS frameworks, with potential for deployment in enterprise, cloud, and edge environments.