About reliability estimation of detectors and Bayesian reasoning

PDF

Authors
  1. Jousselme, A-L.
Corporate Authors
Defence Research and Development Canada, Valcartier Research Centre, Quebec QC (CAN)
Abstract
This brief note addresses the problem of performance estimation of detectors, with a specific emphasis on detectors for rare events. The discussion is framed in the Bayesian reasoning framework for estimating posterior probability P(E|A) about events E based on observations A. The question addressed is which measure between P(A|E) or P(E|A) should be used as a requirement for performance of detectors. The two measures are first presented and their respective meaning discussed. Beyond the semantics associated to the term “reliability”, the semantics of the mathematical quantities concerned is explained and the impact of this measure on detector’s designers is discussed. Some ideas to improve performances of detectors for rare events are finally sketched, emphasising that the need for such detectors to be only components of a larger situation assessment system, so that correlations with other detectors or other pieces of information can be drawn, and that the decision maker can decide based on a more global picture of the situation.

Il y a un résumé en français ici.

Keywords
sensors reliability;bayesian reasoning;measures of performance;detection;performances;ROC curve;reliability;false alarms;fusion
Report Number
DRDC-RDDC-2016-R040 — Scientific Report
Date of publication
01 Apr 2016
Number of Pages
28
DSTKIM No
CA042366
CANDIS No
803633
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: