Classification methods multi-class multivariate longitudinal data

Abstract

In this thesis, we have focussed on the classification of multiple class, multivariate longitudinal data. This research was driven by a study conducted to classify psychotropic drugs based on electro-encephalogram or EEG data. For each of the compound-by-dose combinations in the five psychotropic drug classes, data on the sleep-wake behaviour of rats were collected during a 16 hours period. The sleepwake behaviour was summarized into six standard sleep-wake stages, resulting in a six-variate longitudinal profile per rat. From a statistical point of view, analyzing EEG data poses important challenge, because of the high-dimensionality and the longitudinal character of the data. The longitudinal profiles are usually highly irregular and the variability between and within subjects are relatively high.