Performance Analysis of AI-Enhanced Cybersecurity Networks

Abstract

The sophistication of cyberattacks is on the rise and it demands sophisticated and dynamic intrusion detection systems that will help to detect both the existing and the new threats. This paper is a performance appraisal of intrusion detection models based on artificial intelligence with two current benchmark datasets, namely, UNSW-NB15 and CSE-CIC-IDS2018. It compares classical machine learning techniques with deep learning models, such as Convolutional Neural Networks and Gated Recurrent Units with attention mechanisms and suggests a new hybrid model, consisting of Stacked Autoencoders in combination with Gradient Boosting to learn features and classify better. Strict experimental regimen using leakage-free preprocessing, cross-validation and multi-metric evaluation was used to ensure fairness and reproducibility. The experimental findings prove almost flawless performance in terms of detection in both datasets, with the hybrid Autoencoder-Gradient Boosting model and the attention-based recurrent network being stronger in the face of the imbalance of classes and sophisticated traffic patterns. Besides being highly accurate, the proposed approach is more stable, generalized, and interpretable, which is why it is appropriate to deploy it in practice. The results affirm the fact that hybrid artificial intelligence systems are capable of improving cybersecurity defenses. Further research is needed on scalable, privacy-preserving, and interpretable intrusion detection on new network conditions emerging e.g. Internet of Things and edge computing systems.