A Comparative Study of K-means and NMF Clustering Algorithms

Cluster analysis, also known as clustering, is a multivariate data mining technique whose goal is to classify objects based on a set of selective features. Nonnegative matrix factorization (NMF), and K-means are the most used clustering methods. NMF is a method that breaks down a high-dimensional data matrix into two lower-dimensional matrices with positive elements; while K-means is an algorithm that assigns each data point to a cluster that has a centroid. The main objective of this algorithm is to minimize the total distance between the data points and their respective clusters. In this paper, four datasets, consist of data collected from newswires, radio programs, and television programs are prepared for assessing. Then, they are clustered via NMF and K-means techniques. Furthermore, the results are compared with Purity and Entropy criteria. It can be concluded that both algorithms work better in small dimension datasets rather than high-dimensional ones. Furthermore, NMF has generally a better accuracy.