Using the mean distance criterion to identify novelty in the data

Abstract

Using the mean distance criterion to identify novelty in the data

Krasheninnikov A.M.

Incoming article date: 02.08.2021

The article discusses the features of identifying novelty in data, as well as general methods for identifying it. Since the absence of noise in the training information is a determining factor for building high-quality classifiers on it in supervised machine learning, such a practically important special case of the search for novelty is considered, when it is determined in separate classes of training data after all outliers have been eliminated in these data. For greater definiteness, when searching for novelty, its geometric interpretation in the space of object feature values.

Keywords: data, classifier, outliers, novelty, novelty detection, geometric approach, statistical criterion