A Robust Method in Detecting Change-Points

• Activity Number: 655
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Yi Liu*+ and David Siegmund and Nancy Zhang
• Companies: Stanford University and Stanford University and Stanford University
• Keywords: change-point ; copy number variation ; robust ; non parametric

Abstract

Change-point models have been widely applied for segmenting spatial or time-series data. One common assumption in these models is that observations follow Gaussian distribution. If the errors have heavy tails, which is common in copy number data, a change-point model based on a normality assumption will suffer from high rate of false positives. Motivated by this, we propose a non-parametric change-point model for segmentation and give approximations for the significance level of the test. We show that our method is more robust when errors have heavy tails, and loses little power even if the Gaussian assumption is true.