DOI: https://doi.org/10.36347/sjet.2026.v14i03.003

Pages: 143-152

Multi-tenant cloud environments are vulnerable to the security issues of dynamic traffic behaviors, encrypted communication, and the vulnerability of shared resources, which restricts the performance of traditional intrusion detection systems. This paper presents a proposal of an Intelligent Cloud Intrusion Detection Framework (ICIDF) based on a hybrid deep-learning system that combines the Convolutional Neural Networks to learn the hierarchical features, with Long Short-Term Memory networks to learn how to model the temporal dynamics of traffic. The pipeline of Recursive Feature Elimination-Principal Component Analysis was utilized in full exhaustively to minimize the dimensionality of features to improve the computation speed. The model was tested on the NSL-KDD, CICIDS2017 and CSE-CIC-IDS2018 datasets and then tested in an actual real-time OpenStack-based multi-tenant cloud infrastructure. The model proposed had an accuracy of up to 99.02, an F1 -score of 98.90 and an AUC of 0.998, a false-positive rate of 0.92, thus doing better than traditional machine-learning methods, individual deep-learning methods, and autoencoders. The dimensionality reduction of feature optimization was 69.2% and this allowed quicker convergence and scalability to deployment. Liveness testing with real time showed a 2.8ms detection latency, 18 400 flows/s throughput, and less resource overhead with a trivial performance degradation as the number of tenants grew. These findings suggest that joint spatial-temporal deep-representation learning offers accurate, scalable and real-time intrusion detection of multi-tenant cloud infrastructures. The suggested framework provides a viable basis of SLA-conscious, autonomous cloud security and massive scale deployment.