Traffic Abnormality Detection ML-based Method for IoT

Network traffic monitoring is an vital questions for the networks security and reliability. The statistical model of traffic is at the heart of many methods for detecting traffic anomalies. Existing modern methods of detecting attacks in several cases turn out to be insufficiently reliable, for example, due to the missed moment of the attack, which makes it possible for an attacker to introduce errors into the operation of the system and make it unusable (for example, to carry out a DDOS attack). The paper describes the procedure for reducing the feature space of the original dataset, which made it possible to significantly reduce its dimension. The focus of the research is to use of statistics, correlation and information parameters of IoT traffic for ML-based traffic abnormality detection. The quality of the ten most common machine learning models was assessed on the obtained preprocessed subsample of data.