Cybersecurity in Cloud Computing AI-Driven Intrusion Detection and Mitigation Strategies

This paper explores the development and evaluation of advanced machine learning models for intrusion detection in cloud environments. We focus on Transformer-based Spatio-Temporal Graph Neural Networks (ST-GNN), CNN, LSTM, Isolation Forest, and conventional GNNs, analyzing their performance on three distinct datasets: NSL-KDD, CICIDS2017, and a custom synthetic dataset. The models were assessed based on key metrics such as precision, recall, F1 score, ROC-AUC, and detection latency. Our results demonstrate that Transformer-based ST-GNN exhibits superior performance, showing robustness, scalability, and efficient real-time detection capabilities, making it a promising candidate for next-generation intrusion detection systems (IDS). We also discuss the mathematical foundations behind model superiority, including generalization bounds, and acknowledge the limitations of current models, such as vulnerability to adversarial attacks. The study highlights the potential for improvements in real-time federated deployment, hardware-aware acceleration through FPGA/GPU-based inference, and integration with Zero-Trust Architecture (ZTA) for enhanced cybersecurity. This paper provides a comprehensive comparison of IDS models, offering valuable insights for future research and real-world applications in network security.