ROC Analysis of Extreme Seeking Entropy for Trend Change Detection

Abstract

This paper is dedicated to the evaluation of the ROC curve of recently introduced Extreme Seeking Entropy algorithm. The ROC curve is evaluated for a trend change in the signal that contains additive Gaussian noise. The resulting ROC curve of the Extreme Seeking Entropy algorithm is compared with other adaptive novelty detection methods, namely Learning Entropy and Error and Learning Based Novelty Detection as those algorithms are also evaluating the adaptive weights increments. The ROC curves are evaluated for multiple noise variances and area under those ROC curves is estimated.