Classification for Repeated Measures Data using Data Mining Methods
by
Denny Meyer
Institute of Information and Mathematical Sciences, Albany Campus, Massey

Neural networks have been used to successfully forecast time series data and to classify observations according to group membership. This paper investigates the performance of neural networks in the context of repeated measures data. In particular, the effect of autocorrelation strength and the effect of the length of the time series are studied using simulated data. Real world data sets are used to compare the classification efficiency of neural networks with that of classification trees and conventional discriminant analysis methods.

Date received: May 26, 2002