Feature Engineering for Deep Learning-Based Anomaly Detection in 5G and Beyond

In the current digital landscape characterized by an immense number of interconnected devices, the possibilities for potential network attacks are virtually limitless. This issue is further exacerbated by the fact that numerous low-cost devices lack robust security measures, rendering them susceptible to hacking. Once compromised, these devices can be enlisted in a botnet to execute distributed denial-of-service (DDoS) attacks. In response to this escalating cybersecurity threat, we propose an innovative Intrusion Detection System (IDS) that capitalizes on deep learning, utilizing optimal feature extraction from network traffic. By trimming down the number of features by a substantial 94% without sacrificing classification accuracy, our method demonstrates resilience to minor system perturbations that do not signify a network anomaly. Our approach underwent rigorous evaluation in diverse simulation scenarios, delivering a high detection accuracy of 99% on well-established datasets. Moreover, it manages to reduce the training time by an order of magnitude.