MPEG VBR video traffic classification using Bayesian and nearest neighbor classifiers

Abstract

We propose a Bayesian classifier and a nearest neighbor classifier (NNC) for MPEG variable bit rate (VBR) video traffic based on the I/P/B frame sizes. Our simulation results show that: 1) MPEG video traffic can be classified based on the I/P/B frame sizes only using the Bayesian or nearest neighbor classifiers, and both classifiers can achieve quite low false alarm rate; 2) the nearest neighbor classifier performs better than the Bayesian classifier which seems ridiculous because the Bayesian classifier is recognized as the optimal classifier. The reason is because the recognized log normal distribution is not a good approximation for I/P/B frame sizes. The Bayesian classifier is a model-based classifier (based on the log normal distribution in this paper), and the nearest neighbor classifier is model free, so it can perform better than the Bayesian classifier.