Table 3. Prediction accuracies using ten times 10-fold cross validation with different gene subsets.

Data Used	Learning Models
	SVM	RF	KNN	DLDA
All Genes (10592 genes)	45.9%±5.7	52.9%±11.3	61.1%±8.1	52.6%±8.2
Fisher Analysis using a Training Set (Top 100 genes)	49.3%±11.3	53.4%±7.6	53.0%±6.6	54.2%±5.4
Student's T-Test Analysis using a Training Set (Top 100 genes)	45.7%±10.5	51.6%±5.7	47.9%±13.9	56.6%±6.1
SAM Analysis with an FDR = 8% (34 genes)	70.2%±5.7	74.9%±8.1	73.7%±4.5	80.9%±2.2
Fisher Analysis (Top 100 genes)	67.8%±11.2	69.8%±4.6	54.9%±6.3	70.6%±4.0
Student's T-Test Analysis (Top 100 genes)	77.8%±7.9	71.0%±4.9	74.0%±4.9	78.8%±2.8
Combined SAM, Fisher, & Student's T-Test (9 genes)	73.0%±7.1	68.5%±5.3	66.5%±4.0	69.2%±3.2
Golub Cancer Data (7129 genes)	98.6%±0.0	96.9%±2.2	92.0%±2.9	89.0%±4.1

SVM, support vector machine; RF, random forest; KNN, K-nearest neighbour, DLDA, diagonal linear discriminant analysis; SAM, significance analysis of microarrays; FDR, false discovery rate. Values represent mean±standard deviation corresponding to the 95% confidence interval.