Functional Variance Change Point Analysis for Big Data with an Application to Liver Procurement

• Activity Number: 570 - New Frontiers of Functional Data Analysis
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Pang Du* and Zhenguo Gao and Ran Jin and John Robertson
• Companies: Virginia Tech and Virginia Tech and Virginia Tech and Virginia Tech
• Keywords: Functional variance change point; Thin-plate splines; Subsampling; Functional data; Liver procurement

Abstract

Liver procurement via surface-temperature monitoring calls for development of a variance change point detection method under a smoothly-changing mean trend. Gao et al (2018) considered a spot-wise approach and establish the theoretical foundation of such change point analysis methods. But it does not offer immediate information to surgeons since an organ is often transplanted as a whole or in part. Therefore we develop a new method that can analyze a chunk of the organ surface at a time. Besides its practical appealingness, our method provides a novel addition to the developing field of functional data monitoring. Also we need to resolve the numerical challenge of simultaneously modeling the variance functions of 2D location and the mean function of location and time with sample sizes in the scales of 10000 and 1M respectively. Standard spline estimation would be too costly or even impossible at such scales. So we introduce a multi-stage subsampling strategy consisting of several down-sampling or subsampling steps educated by cheap preliminary statistical measures. Extensive simulations and application to data from the liver procurement experiment are provided.