Table 12

Comparison of best classification accuracy for the Prostate Cancer dataset

Methods (feature selection + classification)

#Selected genes

#Correctly classified samples (accuracy)

Rule-based classifier

depended degree + decision rules [this work]

1

31 (91.18%)

yes

2

27 (79.41%)

TSP [14]

2

32 (94.12%)

yes

PCLs [50]

unknown

33 (97.06%)

yes

discretization + Single C4.5 [11]

unknown

23 (67.65%)

yes

discretization + Bagging C4.5 [11]

unknown

25 (73.53%)

yes

discretization + AdaBoost C4.5 [11]

unknown

23 (67.65%)

yes

RCBT [13]

unknown

33 (97.06%)

yes

SVMs [13]

unknown

27 (79.41%)

no

signal to noise ratios + k-NNs [18]^d

4

26 (77.2%)

no

16

29 (85.7%)

no

^dIn [18], as both raw and normalized datasets were used, two groups of prediction results were obtained. Here, we chose their results from the normalized dataset. Another small difference is that we obtained the dataset from the Kent Ridge Bio-medical Data Set Repository, where the prostate test set includes 25 tumor and 9 normal samples instead of the 27 tumor and 8 normal samples studied in [69]. To facilitate comparison, the correctly classified sample numbers were calculated according to the total of 34 samples.

Wang and Gotoh BMC Medical Genomics 2009 2:64   doi:10.1186/1755-8794-2-64