Dynamic Classification Using Multivariate Locally Stationary Wavelets

• Activity Number: 388
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Timothy Park*+ and Idris Eckley and Hernando Ombao
• Companies: Lancaster University and Lancaster University and University of California, Irvine
• Keywords: classification ; wavelets ; nonstationary time series ; multivariate time series

Abstract

Methods for the supervised classification of time series generally aim to assign a series to one class for its entire time span. In this paper we present an alternative formulation where the class membership of a series is permitted to change over time. Our aim therefore changes from classifying the time series as a whole to classifying the series at a particular time point. We assume that each class is characterised by a different stationary generating process, the series as a whole will however be nonstationary due to class switching. To capture this nonstationarity requires a nonstationary model. Here we use the Multivariate Locally Stationary Wavelet model which is particularly suitable as it captures the nonstationary dependence structure of a series, this will then be used for classification. In order to account for uncertainty in class membership our goal is not to assign a definite class membership at a particular time point but rather to calculate the probability of a series belonging to a particular class at a particular time point. This method is shown to perform well with both simulated and real EEG data.