Deep Learning-based Intelligent Threat Detection using Multi-Source Cybersecurity Data and Advanced Feature Engineering

Abstract

As sophisticated and diverse attacks and security data continue to grow at an alarming rate, the need for intelligent threat detection systems that can accurately detect the new patterns of attack has grown. Most of the traditional cybersecurity strategies are based on the labeled data and are limited to security sources. In order to utilize the advantages of deep learning and intelligent threat detection in multi-source cybersecurity data with advanced feature engineering, this work suggests an Adaptive Graph Contrastive Feature Learning (AGCFL) framework. The proposed solution combines network traffic, endpoint telemetry, cloud logs, intrusion detection system logs, and threat intelligence by leveraging a graph-based approach for constructing sessions, temporal encoding of the features, self-supervised contrastive learning, and attention-based feature selection. Feature representations are optimized, and stored in a security feature repository to facilitate quick threat classification and ongoing learning. Experimental results show that AGCFL can outperform the recently developed state-of-the-art algorithms in terms of detection accuracy, false positive rate and computational efficiency. The proposed framework will enable enterprises to have a scalable, adaptive, and robust cybersecurity solution for enterprise networks, cloud infrastructure, and intelligent security operation centers.