Performance Evaluation of AI Driven Cybersecurity Intrusion Detection Systems Using Adversarial Traffic in Encrypted Networks

Abstract

The growth of encrypted network traffic in recent years has made it much more difficult to identify advanced cyberthreats. In these kinds of settings, traditional intrusion detection systems (IDS) frequently find it difficult to remain accurate, especially when confronted with maliciously constructed traffic that is intended to avoid detection. The performance of AI driven cybersecurity intrusion detection systems running inside encrypted network infrastructures is thoroughly assessed in this study.In order to replicate realistic adversarial scenarios, a controlled testbed was created that included datasets with both malicious and benign encrypted flows. Under various degrees of adversarial perturbations, a number of machine learning and deep learning models, such as Random Forest, Support Vector Machine, and Convolutional Neural Networks, were trained and assessed. Performance metrics such as accuracy, precision, recall, F1 score, and ROC-AUC were measured to quantify detection capability. The results demonstrate that while AI driven IDS significantly outperform traditional signature based methods, their resilience decreases under high intensity adversarial traffic, particularly in scenarios with limited feature visibility due to encryption. This research highlights the importance of incorporating adversarial training, feature engineering, and adaptive learning strategies to enhance IDS robustness in encrypted environments. The findings provide actionable insights for the development of next generation cybersecurity solutions capable of mitigating advanced evasion techniques.