Non-negative Matrix Factorization and Classifiers: Experimental Study

O.G. Okun (Finland)


Dimensionality reduction, pattern classification, machine learning.


Non-negative matrix factorization (NMF) is one of the re cently emerged dimensionality reduction methods. Unlike other methods, NMF is based on non-negative constraints, which allow to learn parts from objects. In this paper, we combine NMF with four classifiers (nearest neighbor, ker nel nearest neighbor, k-local hyperplane distance nearest neighbor and support vector machine) in order to investi gate the influence of dimensionality reduction performed by NMF on the accuracy rate of the classifiers and estab lish when NMF is useless. Experiments were conducted on three real-world datasets (Japanese Female Facial Ex pression, UCI Sonar and UCI BUPA liver disorder). The first dataset contains face images as patterns whereas pat terns in two others are composed of numerical measure ments not constituting any real physical objects when as sembled together. The preliminary conclusion is that while NMF turned out to be useful for lowering a dimensionality of face images, it caused a degradation in classification ac curacy when applied to other two datasets. It indicates that NMF may not be good for datasets where patterns cannot be decomposed into meaningful parts, though this thought requires further, more detailed, exploration. As for classi fiers, k-local hyperplane distance nearest neighbor demon strated a very good performance, often outperformingother tested classifiers.

Important Links:

Go Back