Changepoint detection for dependent Gaussian sequences

In this paper techniques are devised for detecting changepoints in Gaussian sequences, with the distinguishing feature that we do not impose the assumption that the series' terms be independent. For the specific case that the Gaussian sequence has a creation structure of ARMA type, we develop CUSUM-like procedures; we do so by relying on a large-deviations based approach. In the networking context, these tests can for instance be used to detect a change in traffic parameters, such as the mean, variance or correlation structure. The procedures are extensively validated by means of a broad set of simulation experiments.