Archetypal Curves in the Shape and Size Space: Discovering the Salient Features of Curved Big Data by Representative Extremes

Abstract

Curves are complex data. Tools for visualizing, exploring, and discovering the structure of a data set of curves are valuable. In this paper, we propose a scalable methodology to solve this challenge. On the one hand, we consider two distances in the shape and size space, one well-known distance and another recently proposed, which differentiate the contribution in shape and in size of the elements considered to compute the distance. On the other hand, we use archetypoid analysis (ADA) for the first time in elastic shape analysis. ADA is a recent technique in unsupervised statistical learning, whose objective is to find a set of archetypal observations (curves in this case), in such a way that we can describe the data set as convex combinations of these archetypal curves. This makes interpretation easy, even for non-experts. Archetypal curves or pure types are extreme cases, which also facilitates human understanding. The methodology is illustrated with a simulated data set and applied to a real problem. It is important to know the distribution of foot shapes to design suitable footwear that accommodates the population. For this purpose, we apply our proposed methodology to a real data set composed of foot contours from the adult Spanish population.