With ever-increasing numbers of astrophysical transient surveys, new facilities and archives of astronomical time series, time domain astronomy is emerging as a mainstream discipline. However, the sheer volume of data alone - hundreds of observations for hundreds of millions of sources – necessitates advanced statistical and machine learning methodologies for scientific discovery: characterization, categorization, and classification. Whilst these techniques are slowly entering the astronomer’s toolkit, their application to astronomical problems is not without its issues. In this paper, we will review some of the challenges posed by trying to identify variable stars in large data collections, including appropriate feature representations, dealing with uncertainties, establishing ground truths, and simple discrete classes.