A method of outliers detection based on amend sequential probabilistic ratio analysis

Outlier detection is a statistical problem that has received considerable attention. A common approach is assuming that the (possible) outliers are generated by contaminating models. It is known that sequential probabilistic ratio analysis (ASPR) is not very sensitive to outliers. Therefore, identification of outliers is possible for exploring appropriate model structures and determining reliable estimates of parameters. This paper examines the use of amend sequential probabilistic ratio analysis for outlier detection. We develop identification indices for detecting observations that influence the SPR estimates, higher. Finally, an example is given to illustrate the daily average number of car manufacturing defects application in the proposed detection.