Abstract
Let $\mathbf{X}=(X_1,X_2,\ldots,X_n)$ be a time series, that is a sequence of random variable indexed by the time $ t=1,2,\ldots,n $. We assume the existence of a segmentation $\tau=(\tau_1,\tau_2,\ldots,\tau_n)$ such that $X_i$ is a family of independent identically distributed (i.i.d) random variable for i $\in (\tau_k,\tau_k+1],~and~k=0,\ldots,K$ where by convention $\tau_o$ and $\tau_{K+1}=N$. In the literature, it exist two main kinds of change points detections: The change points on-line and the change points off-line. In this work, we consider only the change point analysis (off-line), when number of change points is unknown. The result obtained is based on Filtered Derivative method where we use a second step based on False Discovery Rate. We compare numerically this new method with the Filtered Derivative with p-Value.
Highlights
Change-point detection is an important problem in many applications, and it has been well-studied for a long time, see e.g. the textbooks (Basseville & Nikirov, 1993; Brodsky & Darkhovsky, 1993; Csorgo & Horvath, 1997), or (Huskova & Meintanis, 2006b; Gombay & Serban, 2009) for an updated overview
We propose to replace the family of single hypothesis tests of Step 2 in Filtered Derivative with p-value (FDpV) method by the use of the False Discovery Rate
We propose a new method derived from the FDpV one: We replace Step 2 of FDpV by False Discovery Rate method (FDR) and we call this method FDqV
Summary
Change-point detection is an important problem in many applications, and it has been well-studied for a long time, see e.g. the textbooks (Basseville & Nikirov, 1993; Brodsky & Darkhovsky, 1993; Csorgo & Horvath, 1997), or (Huskova & Meintanis, 2006b; Gombay & Serban, 2009) for an updated overview. We only consider the a posteriori problem In this century, the state to the art method was the Penalized Least Square Criterion (PLS): When the number of change point is known, PLS minimizes a contrast function (Bai & Perron, 1998; Lavielle & Moulines, 2000). Cumulative sum can be iteratively computed and leads to algorithms with both time and memory complexity of order O(n) Among these methods, the Filtered Derivative has been introduced by Benveniste and Basseville (1984) and Basseville and Nikirov (1993). We propose to replace the family of single hypothesis tests of Step 2 in FDpV method by the use of the False Discovery Rate.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
More From: International Journal of Statistics and Probability
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.