SRSaRa: A SaRa-Inspired Modification of Pettitt's Test for Non-Parametric Change-Point Detection

Abstract

The Signed-Rank Screening and Ranking Algorithm or SRSaRa is a non-parametric change-point detection technique that is based on a SaRa-like process with a diagnostic function inspired by Pettitt's test. Possessing two modes, `LM' and `MAX' for single and multiple change-point detection respectively, the SRSaRa is flexible and robust to outliers through its diagnostic function. The SRSaRa's `MAX' mode for single change-point detection outperforms Pettitt's test in several scenarios while maintaining Type-I error control, while the SRSaRa's `LM' mode is capable of controlling FDR at the desired level and shows promise as a non-parametric multiple change-point detection technique.