Contrastive Autoencoding with Gaussian Confidence Regions for Concept Drift Detection in IDS

Intrusion Detection Systems (IDS) are essential for network security; however, the growing complexity of cyberattacks challenges traditional signature-based and anomaly-based approaches, which struggle to detect novel threats while maintaining low false positive rates. In dynamic environments, evolving attack strategies cause concept drift that degrade the performance of static models. To address this, we propose a novel machine learning approach that integrates an autoencoder with contrastive learning and models known attack classes using Gaussian-based confidence regions. Experimental results show that the proposed classifier outperforms the baseline approach, achieving a higher average F1-score (0.39 vs. 0.25) due to the adaptability of hyperellipsoidal confidence regions.