Performance Examination of Hard Clustering Algorithm with Distance Metrics

Abstract

Clustering algorithms based on partitions are widely us ed in unsupervised data analysis. K-means algorithm is one the efficient partition based algorithms ascribable to its intelligibility in computational cost. Distance metric has a noteworthy role in the efficiency of any clustering algorithm. In this work, K-means algorithms with three distance metrics, Hausdorff, Chebyshev and cosine distance metrics are implemented from the UC Irvine ml-database on three well-known physical-world data-files, thyroid, wine and liver diagnosis. The classification performance is evaluated and compared based on the clustering output validation and using popular Adjusted Rand and Fowlkes-Mallows indices compared to the repository results. The experimental results reported that the algorithm with Hausdorff distance metric outperforms the algorithm with Chebyshev and cosine distance metrics.