Title: Improved scale-space analysis for interval-valued data
Authors: Kee-Hoon Kang - Hankuk University of Foreign Studies (Korea, South) [presenting]
Cheolwoo Park - University of Georgia (United States)
Yongho Jeon - Yonsei University (Korea, South)
Abstract: With the rapid advancement of computing technology and storage capacity, both the size of data and the complexity of their structure have significantly increased. These enormous data are sometimes converted into new types of data such as intervals, histogram, and trees, etc. Interval-valued data represent uncertainty or variability and offer richer and more complex information on trend and variation of the underlying structure than single-valued data. One can sample single-valued data from interval-valued data by assuming a uniform distribution. This can be improved if the single-valued data are generated from the actual underlying distribution rather than a uniform distribution. Recently, a nonparametric method of estimating a joint density using the interval-valued data is developed. It treats the observed set of hyper-rectangles as a multivariate histogram that can be approximated to locally weighted Gaussian kernel functions. We consider this approach with SiZer map for interval-valued data in a nonparametric regression setting.