| Summary: | Thesis (M.Sc.)--Memorial University of Newfoundland, 2011. Mathematics and Statistics Bibliography: leaves 82-85. The identification of changes in process parameters is an important statistical problem in industrial-process monitoring. The existing methods, the change point model (Hawkins et al. (2003)) and the modified information criterion (Chen et al. (2006)) rely on the parametric distribution of the quality characteristic, and any deviation from the specified model may lead to incorrect conclusions. We propose an empirical-likelihood-based information criterion (ELIC) for identifying changes in the process parameters. The main advantage of our method is that we do not need to specify a parametric distribution for the quality characteristic. Our simulation studies indicate that our method is as good as existing methods when the distribution of the quality characteristic is known, and it outperforms existing methods when the distribution is approximated or misspecified. We introduce the EM test in the Bayesian approach for the change point problem suggested by Bansal et al. (2008). From simulation studies, we see that the Bayesian EM test performs as well as the Bayesian approach with full EM iteration. We compare the performance of all methods for identifying the change point in a wide range of data scenarios. Our methods are applied to two case studies.
|
|---|