Junghye Lee a,* , In Young Choi b , Chi-Hyuck Jun c
a Department of Industrial Engineering, Ulsan National Institute of Science and Technology (UNIST), 50 UNIST-gil, Ulsan, Republic of Korea
b Department of Medical Informatics, College of Medicine, The Catholic University of Korea, 222 Banpo-daero, Seoul, Republic of Korea
c Department of Industrial and Management Engineering, Pohang University of Science and Technology (POSTECH), 77 Cheongam-ro, Pohang, Republic of Korea

Abstract

Classification of microarray data plays a significant role in the diagnosis and prediction of cancer. However, its

high-dimensionality (>tens of thousands) compared to the number of observations (<tens of hundreds) may lead

to poor classification accuracy. In addition, only a fraction of genes is really important for the classification of a

certain cancer, and thus feature selection is very essential in this field. Due to the time and memory burden for

processing the high-dimensional data, univariate feature ranking methods are widely-used in gene selection.

However, most of them are not that accurate because they only consider the relevance of features to the target

without considering the redundancy among features. In this study, we propose a novel multivariate feature

ranking method to improve the quality of gene selection and ultimately to improve the accuracy of microarray

data classification. The method can be efficiently applied to high-dimensional microarray data. We embedded the

formal definition of relevance into a Markov blanket (MB) to create a new feature ranking method. Using a few

microarray datasets, we demonstrated the practicability of MB-based feature ranking having high accuracy and

good efficiency. The method outperformed commonly-used univariate ranking methods and also yielded the

better result even compared with the other multivariate feature ranking method due to the advantage of data

efficiency.