Identifying features with concept drift in multidimensional data using statistical tests

Abstract

Concept drift is a common problem in the data streams, which makes the classifiers no longer valid. In the multidimensional data, this problem becomes difficult to tackle. This paper examines the possibilities of identifying the specific features, in which concept drift occurs. This allows to limit the scope of the necessary update in the classification system. As a tool, we select a popular Kolmogorov-Smirnov test statistic.