Skip to main content

Table 2 Comparison of R-SVM and SVM-RFE on Data-S (with sample outliers)

From: Recursive SVM feature selection and sample classification for mass-spectrometry and microarray data

Levela

ReduceSVb

P(sv-diff)c

ReduceTestd

P(test-diff)e

ImproveRecf

P(rec-diff)g

ReduceOSVh

P(osv-diff)i

800

3.25%

4.49E-41

-65.19%

5.65E-36

-10.14%

3.36E-75

50.37%

5.97E-35

600

5.80%

1.90E-57

-70.27%

3.04E-35

-7.14%

5.18E-56

72.28%

1.10E-49

500

7.02%

8.20E-63

-59.63%

1.81E-37

-5.13%

3.37E-39

80.54%

1.17E-56

400

8.26%

1.68E-67

-41.43%

8.31E-25

-2.57%

4.53E-12

89.04%

2.51E-64

300

7.72%

1.20E-58

-19.14%

2.18E-13

0.75%

4.92E-02

93.44%

7.46E-65

200

7.21%

4.54E-51

-6.53%

2.56E-04

4.00%

7.15E-16

93.91%

1.47E-61

150

9.13%

1.29E-71

2.63%

1.20E-01

6.47%

8.41E-23

93.59%

6.27E-61

100

8.30%

1.42E-64

5.56%

8.04E-04

7.69%

3.50E-22

92.44%

1.33E-61

90

8.36%

2.01E-72

4.31%

1.15E-02

6.99%

8.74E-19

91.37%

2.60E-61

80

8.01%

6.63E-71

4.45%

1.99E-02

6.99%

9.33E-18

90.26%

2.65E-60

70

7.17%

1.29E-67

6.59%

3.78E-04

7.52%

2.80E-16

88.56%

7.55E-62

60

6.67%

2.65E-65

6.16%

2.32E-03

7.27%

5.72E-13

86.38%

2.60E-62

50

5.82%

1.08E-58

7.70%

1.34E-04

7.42%

3.71E-12

83.82%

1.23E-61

  1. a,b,c,d,e,f,g same as in Table 1.
  2. h ReduceOS V: Relative reduction in the number of outlier support vectors (the outlier samples being taken as support vectors) in R-SVM comparing to that in SVM-RFE, calculated as: (average #OSVSVM-RFE - average #OSVR-SVM)/(average #OSVSVM-RFE), where #OSV denotes the number of outlier samples being taken as support vectors by the method mentioned in subscript.
  3. i P(osv-diff): The p-value of observed difference in OVS, by paired t-test.