Application of genetic algorithms and constructive neural networks for the analysis of microarray cancer data

Table 7 Differences between feature selection algorithms

Classifier	Dataset	p-value	Control	Statistically different FS procedures
LDA	Leukemia	1.54e⁻¹²	SFS	GA
	Lung	1.54e⁻¹²	SFS	GA
	Colon	1.54e⁻¹²	SFS	GA
	Breast	1.54e⁻¹²	SFS	GA
	Ovarian	3.28e⁻¹¹	GA	SFS
	Prostate	1.54e⁻¹²	SFS	GA
SVM	Leukemia	3.65e⁻⁵	SFS	GA
	Lung	1.54e⁻¹²	SFS	GA
	Colon	2.86e⁻⁹	GA	SFS
	Breast	1.54e⁻¹²	SFS	GA
	Ovarian	9.13e−11	SFS	GA
	Prostate	1.54e−12	SFS	GA
NB	Leukemia	4.71e−9	SFS	GA
	Lung	1.54e⁻¹²	SFS	GA
	Colon	1.54e⁻¹²	SFS	GA
	Breast	1.54e⁻¹²	SFS	GA
	Ovarian	0.157	-	-
	Prostate	1.54e⁻¹²	SFS	GA
CM	Leukemia	4.71e⁻⁹	SFS	GA
	Lung	1.54e⁻¹²	SFS	GA
	Colon	1.54e⁻¹²	SFS	GA
	Breast	1.54e⁻¹²	SFS	GA
	Ovarian	0.157	-	-
	Prostate	1.54e⁻¹²	SFS	GA
kNN	Leukemia	1.54e⁻¹²	SFS	GA
	Lung	0.0897	-	-
	Colon	1.54e⁻¹²	SFS	GA
	Breast	1.54e⁻¹²	SFS	GA
	Ovarian	0.6547	-	-
	Prostate	1.54e⁻¹²	SFS	GA
NN	Leukemia	4.71e⁻⁹	SFS	GA
	Lung	1.54e⁻¹²	SFS	GA
	Colon	1.54e⁻¹²	SFS	GA
	Breast	1.54e⁻¹²	SFS	GA
	Ovarian	0.157	-	-
	Prostate	1.54e⁻¹²	SFS	GA

Differences between SFS and GA feature selection algorithms for the six different classification methods used (first column). The lowest performant FS procedure is taken as control group (fourth column) while the last column of the table lists the procedures that lead to statistically significant results (corresponding p-value indicated in the third column)

ISSN: 1742-4682