|
Abstract:
|
Change point analysis (CPA) is an important and widely spread research topic in many fields of applications. While many approaches have been proposed for detecting change points, very little of the literature provides comparisons between methods, let alone computational results to help distinguish between techniques. In this presentation, we discuss and compare between two prominent CPA methods namely, the Cumulative Sum (CUSUM) and Likelihood Ratio. The corresponding hypothesis test, test statistic, asymptotic distribution and algorithm are summarized for each approach. Monte Carlo simulations are carried out to show the power of the test statistic using the two methods. We evaluate both methods on detecting change points in sequence of independent samples from different distributions. Further, we assess the accuracy of the estimated change location(s) relative to the actual change location(s) in order to illustrate the effectiveness of the methods at estimating the break locations.
|
