Performance Evaluation of Adversarial Learning for Anomaly Detection using Mixture Models

With the rapid advancement in wireless technology and with the introduction of next-generation 5G-wireless networks, a huge amount of data is generated and transmitted across computer networks worldwide. As a result, we have seen a continuous increase in new network threats and anomalies that creates a significant research challenge to handle the vulnerability along with data integrity, confidentiality, and reliability. More precisely, wireless networks are considered to be highly vulnerable to advanced persistent threat (APT) actors. In this paper, we evaluate an adversarial mechanism for anomaly detection using different statistical mixture models. Indeed, to prevent attacks like data poisoning and evasion from affecting the performance of anomaly detection systems, adversarial learning has proven to be very effective in many research studies. The performance of the mechanism, in which several recently proposed mixture models are integrated, was evaluated using both NSL-KDD and UNSW-NB15 data sets, in terms of accuracy, Detection Rates (DRs), False Positive Rates (FPRs), and computational time.