Parallel K-Means Clustering with Triangle Inequality

Clustering divides data objects into groups to minimize the variation within each group. This technique is widely used in data mining and other areas of computer science. K-means is a partitional clustering algorithm that produces a fixed number of clusters through an iterative process. The relative simplicity and obvious data parallelism of the K-means algorithm make it an excellent candidate for distributed-memory parallel optimization, particularly as datasets grow beyond the size of a single machine. The triangle inequality, when applied to the K-means algorithm, allows unnecessary distance calculations between data objects and cluster centroids to be avoided. Various parallel implementations of the K-means algorithm exist, but no example comparing a standard parallel implementation to one utilizing the triangle inequality could be located. This paper seeks to fill this gap by presenting experimental results demonstrating the performance of both standard and improved parallel K-means implementations compared to a sequential implementation. keywords: K-means, clustering, data mining, parallel, triangle inequality