Comparative evaluation of set-level techniques in predictive classification of gene expression samples

BMC Bioinformatics

Table 5 Generic feature selection (gene-level)

# Method	# Selected Genes	Accuracy				Avg Subrank
		Median	Avg	σ	Iqr
IG	22	90.2	81.5	18.1	30.7	15.0
IG	228	89.8	82.0	17.9	30.3	14.5
SVM-RFE	228	88.3	82.3	16.7	28.5	16.4
SVM-RFE	22	88.0	82.1	17.2	30.4	16.2

Performance of the baseline classification method equipped with a feature-selection step prior to learning. Features (genes) are ranked by the information gain and SVM-RFE heuristics. The number of selected top-ranking genes (22 and 228, respectively) corresponds to the mean number of unique genes acting in gene sets selected in the 1 and 1:10 (respectively) alternatives of the set-level workflow.

ISSN: 1471-2105