Minimum Spanning Tree Based Clustering Algorithms

We propose two Euclidean minimum spanning tree based clustering algorithms — one a k-constrained, and the other an unconstrained algorithm. Our k-constrained clustering algorithm produces a k-partition of a set of points for any given k. The algorithm constructs a minimum spanning tree of a set of representative points and removes edges that satisfy a predefined criterion. The process is repeated until k clusters are produced. Our unconstrained clustering algorithm partitions a point set into a group of clusters by maximally reducing the overall standard deviation of the edges in the Euclidean minimum spanning tree constructed from a given point set, without prescribing the number of clusters. We present our experimental results comparing our proposed algorithms with k-means and the Expectation-Maximization (EM) algorithm on both artificial data and benchmark data from the UCI repository. We also apply our algorithms to image color clustering and compare them with the standard min...

• Type: Conference
• Year: 2006
• Where: ICTAI
• Authors: Oleksandr Grygorash, Yan Zhou, Zach Jorgensen
• Keywords: Artificial Intelligence, Clustering Algorithm, Euclidean Minimum Spanning Tree, ICTAI 2006, Minimum Spanning Trees