Statistical Flow Classification for the IoT.

Cirillo, Gennaro Passerone, Roberto Posenato, Antonio Rizzon, LucaThe objective of this work is to analyze packet flows and classify them as traffic that belongs to IoT devices or to traditional non-IoT communication. We employ two methods: a clustering approach, which learns directly from the structure of the dataset, and a classification tree, trained with the collected data and evaluated using 10-fold cross validation. The results show that classification trees outperform clustering on all datasets, and achieve high accuracy on both homogeneous simulated and real deployment traffic data.