SPARSE STAR COORDINATES: VISUALIZATION FOR HIGH DIMENSION LOW SAMPLE SIZE

Abstract

The visual analysis of group structures and trends of high-dimensional data is a central topic in many fields, particularly in genomic data analysis. Gene expression data have a small number of observations and a large number of attributes. The traditional statistical methods are not directly applied to analyze for high dimension, low sample size. In this paper, we introduce a new visualization technique approach to visual analytics of high-dimension, low-sample size. We propose a sparse star coordinates visualization technique based on star coordinates that group structures are preserved thanks to the optimal layouts of star coordinate systems on the visual space. The larger star coordinates are more important dimensions in cluster analysis. The sparse star coordinate system attains by ranking the best quality visualization of the order of the dominant attributes to analyze the group structures of the high-dimension, low-sample size data sets. We present our proposed method with quality measurement and attest to the effectiveness of our approach for several real data sets.