Making Time: Pseudo Time-Series for the Temporal Analysis of Cross Section Data

The progression of many biological and medical processes such as disease and development are inherently temporal in nature. However many datasets associated with such processes are from cross-section studies, meaning they provide a snapshot of a particular process across a population, but do not actually contain any temporal information. In this paper we address this by constructing temporal orderings of cross-section data samples using minimum spanning tree methods for weighted graphs. We call these reconstructed orderings pseudo time-series and incorporate them into temporal models such as dynamic Bayesian networks. Results from our preliminary study show that including pseudo temporal information improves classification performance. We conclude by outlining future directions for this research, including considering different methods for time-series construction and other temporal modelling approaches.