In this paper we use an improved version of the maximum expectation(ME) algorithm to parametrise a modulated Markov Poisson process. We have found improvements to the k-means algorithm in previously published algorithms that combine the EM estimation method with the k-means algorithm. We are able to show that our modified method produces better results for the parameter estimation. In addition We have used our method to parametrise network traffic measured at the departmental (Imperial College) backbone and another standard traffic source. Using a 4 state MMPP the modelled traffic captures the inter arrival time distribution well.
Information from pubs.doc.ic.ac.uk/parameter-estimation.