On detection of changes in categorical data

Abstract

We consider situations where the observed data is of categorical type and the underlying parameters are subject to abrupt changes of unpredictable magnitude at unknown points in time. We derive change-point detection schemes based on generalized likelihood ratio tests and develop procedures for their design and analysis. We also discuss problems related to parameter estimation for categorical data in the presence of abrupt changes. We illustrate use of the proposed methodology for fault characterization and monitoring in the semiconductor industry.