Cloud Computing is a rapidly expanding technology that provides a range of online services, such as software, computing resources, and databases. However, its shared, distributed, and virtualized nature also makes it highly vulnerable to security threats, including data breaches, unauthorized access, and various types of cyberattacks that can compromise service availability and data integrity. To overcome these issues, a novel MOUNTain gazelle optimized Deep Learning framework (MOUNT-DL) has been proposed for effective intrusion detection and mitigation. The proposed model integrates variational autoencoder-based feature extraction, Mountain Gazelle Optimization (MGO)-based feature selection, and a Temporal Convolutional Network-based Bidirectional Gated Recurrent Unit (TCN-BiGRU) network for classifying the intrusions accurately. The model classifies four attack types and benign traffic using the CICIDS 2017 dataset. The efficacy of the MOUNT-DL approach is assessed utilizing f1-score, precision, accuracy, and recall. Experimental findings demonstrate that the MOUNT-DL approach achieves an accuracy of 98.02% compared to existing methods such as CNN, Deep-IDS, and EOS-IDS