Table 10

Comparison of best classification accuracy for the Leukemia dataset 1

Methods (feature selection + classification)a

#Selected genes

#Correctly classified samples (accuracy)

Rule-based classifier


depended degree + decision rules [this work]

1

31 (91.18%)

yes


2

34 (100%)


t-test, attribute reduction + decision rules [7]

1

31 (91.18%)

yes


attribute reduction + k-NNs [9]

2

33 (97.06%)

no


rough sets, GAs + k-NNs [10]

9

31 (91.18%)

no


EPs [6]

1

31 (91.18%)

yes


discretization + decision trees [11]b

unknownc

31 (91.18%)

yes


CBF + decision trees [24]

1

31 (91.18%)

yes


TSP [14]

2

31 (91.18%)

yes


RCBT [13]

10-40

31 (91.18%)

yes


neighborhood analysis + weighted voting [2]

50

29 (85.29%)

no


signal to noise ratios + PNNs [23]

50

34 (100%)

no


MAMA [25]

132-549

34 (100%)

no


PLS + LD or QDA [26]

50-1500

28-33 (82.4%-97%)

no


prediction strength + SVMs [27]

25-1000

30-32 (88.2%-94.1%)

no


SVMs [28-30]

8-30

34 (100%)

no


aThe text before "+" states the feature selection method, while that after it states the classification method. The absence of "+" means that the same method was used for both feature selection and classification.

bThe decision trees are also involved in feature selection.

c"unknown" means that no related data are provided in the article.

These explanations apply to the other tables.

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

Open Data