Validating Synthetic Health Datasets for Longitudinal Clustering

Abstract

Clustering methods partition datasets into subgroups with some homogeneous properties, with information about the number and particular characteristics of each subgroup unknown a priori. The problem of predicting the number of clusters and quality of each cluster might be overcome by using cluster validation methods. This paper presents such an approach in-corporating quantitative methods for comparison be-tween original and synthetic versions of longitudinal health datasets. The use of the methods is demon-strated by using two different clustering algorithms, K-means and Latent Class Analysis, to perform clus-tering on synthetic data derived from the 45 and Up Study baseline data, from NSW in Australia.