A1195
Title: Nonparametric changepoint detection with theoretically justified thresholds
Authors: Hyeyoung Maeng - Durham University (United Kingdom) [presenting]
Abstract: The purpose is to introduce a nonparametric changepoint detection method for univariate data sequences. In the existing literature, the empirical cumulative distribution function (CDF) is often used to detect changes within a nonparametric framework. However, since the empirical CDF depends on a chosen quantile q, the detection power can vary significantly with this choice. To address this issue, some aggregation techniques for building a test statistic have been proposed, although they often lack theoretically justified thresholds. This could possibly lead to failure in controlling the false positive rate, as the limiting distribution of the test statistic under the null hypothesis is not theoretically justified. Two directions are explored to address this issue: 1) filling this gap by proposing a better threshold based on theoretical justification, and 2) proposing a new cost function built on a set of quantiles, whose exact limiting distribution under the null is known.
