Error sensitivity analysis of Delta divergence - a novel measure for classifier incongruence detection. (May 2018)

• Record Type: Journal Article
• Authors: Kittler, Josef; Zor, Cemre; Kaloskampis, Ioannis; Hicks, Yulia; Wang, Wenwu
• Is Part Of: Pattern recognition. Volume 77(2018:May)
• Page Start: 30
• Page End: 44
• Publication Date: 2018-05
• Subjects: Anomaly detection -- Classifier decision incongruence -- Bayesian surprise; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.11.031
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Theoretical and experimental error sensitivity analysis of Delta divergence as a measure of classifier incongruence. Practical guidelines for selecting classifier incongruence thresholds in practice. Experimental analysis of Delta divergence under varying noise and classi er a posteriori probability distributions. Abstract: The state of classifier incongruence in decision making systems incorporating multiple classifiers is often an indicator of anomaly caused by an unexpected observation or an unusual situation. Its assessment is important as one of the key mechanisms for domain anomaly detection. In this paper, we investigate the sensitivity of Delta divergence, a novel measure of classifier incongruence, to estimation errors. Statistical properties of Delta divergence are analysed both theoretically and experimentally. The results of the analysis provide guidelines on the selection of threshold for classifier incongruence detection based on this measure.