Retraining Mechanism for On-Line Peer-to-Peer Traffic Classification

Abstract

Peer-to-Peer (P2P) detection using machine learning (ML) classification is affected by its training quality and recency. In this paper, a practical retraining mechanism is proposed to retrain an on-line P2P ML classifier with the changes in network traffic behavior. This mechanism evaluates the accuracy of the on-line P2P ML classifier based on the training datasets containing flows labeled by a heuristic based training dataset generator. The on-line P2P ML classifier is retrained if its accuracy falls below a predefined threshold. The proposed system has been evaluated on traces captured from the Universiti Teknologi Malaysia (UTM) campus network between October and November 2011. The overall results shows that the training dataset generation can generate accurate training dataset by classifying P2P flows with high accuracy (98.47%) and low false positive (1.37%). The on-line P2P ML classifier which is built based on J48 algorithm which has been demonstrated to be capable of self-retraining over time.