Efficient Parallelized K-Means for Clustering Big Data

• Activity Number: 460 - Clustering Methods for Big Data Problems
• Type: Topic Contributed
• Date/Time: Wednesday, August 2, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Geoffrey Thompson* and Ranjan Maitra
• Companies: Iowa State University and Iowa State University
• Keywords: k-means ; clustering ; big data

Abstract

Hartigan and Wong's method for k-means clustering has some advantages in both speed and quality of solution over the commonly-used Lloyd's method. However, the latter is readily done in parallel, which makes it feasible to use on large data sets, while the former is not. We present here a parallelized method based on Hartigan and Wong's.