Thinning out redundant empirical data

Abstract

Given a set $X$ of "empirical" points, whose coordinates are perturbed by errors, we analyze whether it contains redundant information, that is whether some of its elements could be represented by a single equivalent point. If this is the case, the empirical information associated to $X$ could be described by fewer points, chosen in a suitable way. We present two different methods to reduce the cardinality of $X$ which compute a new set of points equivalent to the original one, that is representing the same empirical information. Though our algorithms use some basic notions of Cluster Analysis they are specifically designed for "thinning out" redundant data. We include some experimental results which illustrate the practical effectiveness of our methods.