A case study in robust quickest detection for hidden Markov models

We consider the problem of detecting rare events in a real data set with structural interdependencies. The real data set is modeled using hidden Markov models (HMMs), and rare event detection is viewed as a variant of the quickest detection problem. We assess the feasibility of two quickest detection frameworks recently suggested. The first method is based on dynamic programming and follows a Bayesian approach, and the second method is a non-Bayesian approximate cumulative sum (CUSUM) algorithm. We discuss implementation considerations for each method and show their performance through simulations for a real data set. In addition, we examine, through simulations, the robustness of the CUSUM-based method when the rare event model is not exactly known but belongs to a known class of models.