A Hybrid Deep Learning Framework for High-Precision Intrusion Detection in Cloud Infrastructure

With the dynamic nature of cloud computing environments, which involves multiple tenants and large-scale data processing, they are becoming more vulnerable to advanced forms of cyber-attacks. It becomes difficult to detect zero-day attacks, multi-stage intrusions and other evolving threats by traditional Intrusion Detection System (IDS) approaches, which rely on signature and rule-based methods. Even though various studies show that deep learning algorithms like CNN, LSTMs and Transformers perform well, each of them has weaknesses like high false positive rate, lack of robustness against adversarial examples, low explainability and scalability issues. In order to resolve these problems, this paper introduces a new Hybrid Deep Learning IDS Approach combining CNN-based spatial feature extraction, LSTM temporal dependency modeling, Transformer attention mechanism, adversarial training and explainable AI techniques. The main focus is not only to enhance IDS precision but also to make it suitable for real-time implementation within cloud environment. Implementation methods based on benchmark datasets, such as CICIDS2017, CSE-CICIDS2018, UNSW-NB15 and NSL-KDD datasets, are also considered. This work creates a basis for the upcoming implementation and validation steps in further research.