A DEEP LEARNING FRAMEWORK FOR REAL-TIME CYBER THREAT DETECTION AND MITIGATION IN NETWORKED ENVIRONMENTS

The changing nature of the frequency and complexity of cyber threats necessitates the need to implement network security based on intelligent and adaptive security solutions with the ability of detecting and mitigating threats on a real-time basis. This paper proposes an end-to-end framework of cyber threat feature evaluation and mitigation which blends an auto encoder-based hybrid deep learning model, Long Short-Term Memory and Autoencoder (LSTM+AE) with cross correlation-based feature extraction algorithm. The suggested system will dynamically examine incoming network traffic, isolate high-affect features, characterize dangers in real time and apply instant mitigation interventions at transport layer through Transmission Control Protocol (TCP) controls. Context-aware threat attribution techniques are provided by matching network activity to user logs, to improve traceability and responsiveness precision. It has been implemented in Python, with libraries TensorFlow, Keras and Scikit-learn, and the experimental application was tested on the NSK-DD dataset in a virtualized testbed. The experimental accuracy of detection (98.6%), precision (97.9%), recall (98.1%), and the F1-score (98.0%) were high along with the average mitigation latency of less than 1.5 seconds and a rate of false positive of 1.2 percent. Besides, generalizability of the framework was confirmed by protocol-agnostic analysis over TCP, UDP, and ICMP streams. The findings support the system to be effective in improving the resilience of cybersecurity by facilitating pro-active, smart, and efficient management of threats in the complex networked setting.