Skip to main content

Table 11 SVM performances with parameters optimization or not based on informative genes selected by RS

From: Informative gene selection and the direct classification of tumors based on relative simplicity

Parameters optimization

Kernel

Evaluation

Leuk1

Lung1

Leuk2

SRBCT

Breast

Lung2

DLBCL

Cancers

GCM

Average

No (fixed C = 1)

linear

Fitting

97.37

95.31

100

100

100

97.06

100

100

97.92

98.63

LOOCV

97.37

81.25

98.25

98.41

94.44

94.12

96.55

96

77.78

92.69

Testing

94.12

84.38

93.33

95

93.33

97.01

96.67

87.84

58.7

88.93

Yes

linear

Fitting

97.37

100

100

100

100

100

100

99

100

99.6

LOOCV

97.37

90.63

100

100

98.15

94.12

100

96

81.25

95.28

Testing

94.12

84.38

100

95

93.33

95.52

96.67

89.19

65.22

90.38

C

0.25

32

0.03125

0.5

0.125

8

0.25

0.25

4

 

No (fixed C = 1, γ = 1/m)

RBF

Fitting

97.37

87.50

100

100

100

91.18

100

88.00

45.14

89.91

LOOCV

97.37

79.69

100

98.41

98.15

90.44

86.21

78.00

77.08

89.48

Testing

94.12

78.13

100

95.00

93.33

97.01

93.33

85.14

52.17

87.58

Yes

RBF

Fitting

97.37

100

100

100

100

95.59

100

100

100

99.22

LOOCV

97.37

90.63

100.00

98.00

98.15

94.12

100

98.00

82.64

95.43

Testing

94.12

84.38

86.67

90.00

93.33

95.52

90.00

87.84

52.17

86.00

C

8

2048

0.125

0.25

0.5

2

1

32768

32

 

γ

0.0125

0.0075125

0.25

0.125

0.25

0.0625

0.25

0.00390625

0.0625

 
  1. C is penalty parameters and C∈[2−5, 215]; γ is gamma parameter in kernel function and γ∈[2−15, 23]; m is features number of each SVM models