Comparative analysis of methods of unsupervised machine learning for anomaly detection in the IoT systems

Classical unsupervised machine learning methods such as k-nearest neighbor method, histogram-based outlier estimation, isolating forest, cluster local outlier factor are considered in order to identify the most efficient one to be used as the basis for anomaly detection system in IoT traffic. The IoT-23 Dataset, an open-source dataset, is used as the data for the study. The dataset is dimensioned and consists of 23 factors. The study considers an unsupervised learning method «with no teacher» based on copulas, the use of which helps to fully reveal interaction between evaluated factors, which can be successfully used in network traffic analysis in order to detect anomalies. As a result, the accuracy levels of these IoT network anomaly detection methods are compared.