Nonparametric Test for Change-Point Detection in Data Stream

摘要

We offer a new effective online algorithm for implement the Klyushin–Petunin test on streaming data. The Klyushin–Petunin test is a nonparametric test to evaluate the statistical hypothesis that two samples are drawn from the same distribution. The significance level of the test does not exceed 0.05. The test is based on the only assumption that the underlying distribution is absolutely continuous. We present an algorithm for detecting a change in distribution in a data stream. It allows solving two canonical problems of analyzing statistical data — detection of a change-point and detection of a drift of concepts. We show that proposed algorithm is more effective compared to the alternative Kolmogorov–Smirnov and Wilcoxon tests.