Unsupervised Representation Learning for Autonomous Detection of Stealthy Malware and Insider Threats in Encrypted Traffic Streams
Abstract
The rise of encryption protocols has greatly improved data privacy, yet it simultaneously challenges the detection of malicious activities within encrypted traffic. Traditional signature-based techniques struggle to identify stealthy malware and insider threats without decryption. This study proposes an unsupervised representation learning framework to autonomously detect anomalies and threats embedded in encrypted streams. By leveraging autoencoders, contrastive learning, and clustering algorithms, we aim to capture latent patterns indicative of malicious behavior. Experimental evaluations on synthetic and real-world datasets demonstrate that the approach achieves high detection rates with minimal false positives, making it suitable for dynamic and privacy-preserving environments.