Resampling-Based Change Point Estimation

Change point detecting problem is an important task in data mining applications. Standard statistical procedures for change point detection, based on maximum likelihood estimators, are complex and require building of parametric models of data. Instead, the methods of computational statistics replace complex statistical procedures with an amount of computation; so these methods become more popular in practical applications. The paper deals with two resampling tests for change point detection. In well known bootstrap-based CUSUM test we derive the formulas to estimate the efficiency of this procedure by taking an expectation and variance of the estimator into account. We propose also another simple pairwise resampling test and analyze its properties and efficiency too. We illustrate our approach by numerical example considering the problem of decision making of vehicles in city traffic.