Using Deep Packet Inspection Data to Examine Subscribers on the Network
Mike Nkongolo, Jacobus Phillipus van Deventer, Sydney Mambwe Kasongo
Abstract
This article proposes the creation of the deep packet inspection (DPI) dataset to study subscribers’ behavior on the network, applying ensemble learning to this dataset, and comparing it with the UGRansome dataset. The subscriber can be thought of as a person or a group of users using a network service or connectivity. The DPI features represent the subscriber network usage, and the ensemble learning approach is implemented on the DPI dataset to predict the subscriber's service category on the network. The classification and prediction problem addressed on the DPI dataset reached a precision of 100%. The paper predicts that the web and streaming categories with Netflix, Facebook, and YouTube services will be the most utilized in the next few years. This study will lead to a better understanding of the idiosyncratic behavior of active subscribers on the network, exposing novel network anomalies and facilitating the development of novel DPI systems.