Ant based clustering of two-class sets with well categorized objects

Abstract

In the paper, a new ant based algorithm for clustering a set of well categorized objects is shown. A set of well categorized objects is characterized by high similarity of the objects within classes and relatively high dissimilarity of the objects between different classes. The algorithm is based on versions of ant based clustering algorithms proposed earlier by Deneubourg, Lumer and Faieta as well as Handl et al. In our approach, a new local function, formulas for picking and dropping decisions, as well as some additional operations are proposed to adjust the clustering process to specific data. © 2012 Springer-Verlag Berlin Heidelberg.