Title: Pattern recognition techniques for interval time series
Authors: Elizabeth Ann Maharaj - Monash University (Australia) [presenting]
Paula Brito - Universidade do Porto (Portugal)
Paulo Teles - Universidade do Porto (Portugal)
Abstract: An interval time series (ITS) is a sequence of intervals observed in successive instants in time. We focus on the clustering of a set of ITS where we examine existing techniques and propose some new techniques. One new technique involves using a measure that determines the degree of overlap of every pair of ITS under consideration. The measure lies between zero and one and the closer it is to one, the greater the degree of overlap of the two ITS. A distance-type matrix consisting of these measures is used as the input into hierarchical clustering methods. Another new technique involves fitting space-time models to each of the ITS under consideration, and using the parameter estimates of the fitted models as inputs into newly developed and existing clustering methods. Where these techniques differ from existing ones which take into account the upper and lower bounds individually, is that the link between the upper and lower bounds is taken into account.