An Insight on the 'Large G, Small n' Problem in Gene-Expression Microarray Classification

摘要

This paper analyzes the effect of the high-dimensional, low-sample size problem in cancer classification using gene-expression microarrays. Here the two key questions addressed are: (i) What is the percentage of genes that can ensure highly accurate classification?, and (ii) Does this percentage differ from one classifier to another? Both these issues are investigated by developing a pool of experiments with two gene ranking algorithms, five classifiers and four DNA microarray databases.