Change Detection in Teletraffic Models

Abstract

In this paper, we propose a likelihood-based ratio test to detect distributional changes in common teletraffic models. These include traditional models like the Markov modulated Poisson process and processes exhibiting long range dependency, in particular, Gaussian fractional ARIMA processes. A practical approach is also developed for the case where the parameter after the change is unknown. It is noticed that the algorithm is robust enough to detect slight perturbations of the parameter value after the change. A comprehensive set of numerical results including results for the mean detection delay is provided.