Retrospective methods of statistical diagnosis for random processes: ‘contamination’ problems

Abstract

Models of non-stationary random processes considered in this chapter differ substantially from change-points, i.e., abrupt or gradual changes of proba-bilistical characteristics of observations in some intervals of data samples. In ‘contamination’ models of random processes ‘abnormal’ observations are dispersed into the whole sample obtained, while in change-point models they are concentrated in unknown compact areas. Respectively, in ‘contamination’ problems of statistical diagnosis it is required to estimate the share of ‘abnormal’ observations in the whole sample of data and to separate ‘abnormal’ and ‘ordinary’ observations.