This paper provides a solution to evaluate the performance (missed alarm rate, false alarm rate and average detection delay) of alarm systems in which adaptive, dynamic or variable threshold is used.
Confusion matrix has long been used for performance evaluation of machine learning/deep learning-based classification. Deriving accuracy or other performance parameters through using estimated probability density functions can be more advanced and more reliable. However, for complex algorithms, it is not easy to find the separation line (threshold). But what if you can find it? Then you can use our approach to evaluate your ML/DL algorithm in a more advanced way (One step toward Safe AI).
Writing this article was a great pleasure as it has co-authors with whom I have had long-standing collaborations. This article also leads to alarm performance assessment groups contacting me and ultimately to greater involvement in alarm design research.koorosh aslansefat, PhD
ISA Trans 2020 Feb 13;97:282-295. Epub 2019 Aug 13.
MAPNA Electric and Control Engineering and Manufacturing Company, Karaj, Iran. Electronic address:
Download full-text PDF