New Methods of Data Clustering and Classification Based on NMF

Abstract

Nonnegative matrix factorization method is a kind of new matrix decomposition method. It is an effective tool for large data processing and analysis. At the same time, NMF has an important performance on intelligent information processing and pattern recognition. This paper first analyses and discusses the NMF algorithms based on its basic theory. We then propose new methods of data clustering and classification based on NMF separately. NMF method is applied to reduce the dimension of the original matrix. We run clustering algorithms on the encoded matrix after NMF processing instead of on the original matrix. Running clustering algorithms on smaller encoded matrix can save more time and storage space. After that, we bring in a series of improvement methods of classification on the basis of clustering. Finally we have done experiments to test and verify them, and gotten good results.