Evaluation of Machine Learning Algorithms for Anomaly Detection

Malicious attack detection is one of the critical cyber-security challenges in the peer-to-peer smart grid platforms due to the fact that attackers' behaviours change continuously over time. In this paper, we evaluate twelve Machine Learning (ML) algorithms in terms of their ability to detect anomalous behaviours over the networking practice. The evaluation is performed on three publicly available datasets: CICIDS-2017, UNSW-NB15 and the Industrial Control System (ICS) cyber-attack datasets. The experimental work is performed through the ALICE high-performance computing facility at the University of Leicester. Based on these experiments, a comprehensive analysis of the ML algorithms is presented. The evaluation results verify that the Random Forest (RF) algorithm achieves the best performance in terms of accuracy, precision, Recall, F1-Score and Receiver Operating Characteristic (ROC) curves on all these datasets. It is worth pointing out that other algorithms perform closely to RF and that the decision regarding which ML algorithm to select depends on the data produced by the application system.