Publication

Abstract : TCP-based Distributed Denial of Service (DDoS) attacks pose a significant danger to cloud infrastructures because they can imitate genuine traffic patterns, making them difficult to detect using standard approaches. This study introduces the Self-Attention and Intersample Attention Transformer (SAINT) model, a unique deep learning architecture that incorporates Sparse Logistic Regression to address these issues. The SAINT framework uses dual attention mechanisms-self-attention for capturing complicated intraflow dependencies and intersample attention for assessing interflow relationships-to provide enhanced detection of malicious traffic. SAINT, unlike existing methodologies, prioritizes scalability, interpretability, and computational efficiency, distinguishing it from traditional models such as CNNs, RNNs, and ensemble techniques. The model’s efficacy was evaluated using the BCCC-cPacket-Cloud-DDoS-2024 dataset, which included 700,000 traffic flows across 17 advanced attack scenarios, with state-of-the-art metrics: 95% precision, 95% recall, 96% F1 score, and 97% accuracy. Furthermore, studies on the CICDDoS2019 dataset confirmed SAINT’s resilience and flexibility to a variety of network conditions. SAINT addresses real-world issues in cloud-based DDoS detection, such as temporal and spatial traffic complexities, to provide a viable, high performance solution for protecting current cloud infrastructures. This work establishes the groundwork for scalable, adaptable, and efficient cloud-native security frameworks, paving the path for enhanced countermeasures to changing cyber threats.