Discussion on "Sequential detection/isolation of abrupt changes" by Igor V. Nikiforov. Issue 3 (2nd July 2016)

• Record Type: Journal Article
• Authors: Fellouris, Georgios
• Is Part Of: Sequential analysis. Volume 35:Issue 3(2016)
• Page Start: 311
• Page End: 315
• Publication Date: 2016-07-02
• Subjects: Asymptotic optimality -- average run length to false alarm -- changepoint detection -- CUSUM -- false alarm rate -- isolation -- misspecification probability -- worst-case detection delay; 62L10 -- 62L15 -- 60G40; Sequential analysis -- Periodicals; 519.54
• DOI: 10.1080/07474946.2016.1206373
• Languages: English
• ISSNs: 0747-4946

Abstract

ABSTRACT: In a very interesting article, Dr. Nikiforov has reviewed methodological and theoretical developments on the problem of non-Bayesian sequential change detection and isolation. In our discussion of this article, we stress that it is possible to design existing procedures in order to enjoy an asymptotic optimality property (with respect to a Lorden-type criterion) not only under the worst post-change scenario but under every post-change scenario. Moreover, we point out some dangers in working with a more conservative criterion than necessary. Finally, we propose a scalable procedure for the multichannel problem where more than one channels are affected by the change.