Abstract: The current study proposed the development of an automated TIDR system using deep learning to enhance detection and mitigation against network-based cyber threats. For example, traditional signature-based cybersecurity is ineffective in detecting sophisticated attack types such as zero-day exploits and polymorphic malware. In this regard, the paper proposes a deep learning-based approach to spatial and temporal pattern modeling in network traffic by using CNNs, RNNs, transformers, and GNNs. It can be used in real-time threat detection and automated response in an effort to minimize the time required to counteract any emerging threats. Both models returned very high performances; with a malicious activity detection scoring over 98% on the hybrid model CNN-RNN. This research covers limitations concerning the high computational cost and large, labelled dataset requirements of the system and discussing ways to improve in the future using XAI and unsupervised learning. These results will then reflect the fact that the proposed TIDR system will be scalable and adaptive within current cybersecurity challenges and will provide a sound framework for real-time network threat mitigation.