Clustering analysis : models and algorithms / 1192

Abstract

In this paper, the problem of clustering observations into homogeneous groups based on given characteristics of the observations is analyzed.Three distinct integer programming formulations covering important variations of the clustering problem are developed.These variations include finding natural clusters, constraining the number of clusters and restricting the size of clusters.Efficient heuristic techniques employing Lagrangian and eigenvector based methods are developed to solve these problems.