Accurate molecular classification of cancer using simple rules

BMC Medical Genomics

Table 14 Comparison of best classification accuracy for the Leukemia dataset 2

Methods (feature selection + classification)	#Selected genes	#Correctly classified samples (accuracy)	Rule-based classifier
α depended degree + decision rules [this work]	1	14 (93.33%)	yes
HykGene + k-NNs, SVMs, C4.5, NB [85]	26	100%^f	noⁱ
signal to noise ratios + k-NNs [20]	40	95%^g	no
	100	9 (90%)^h

^fLOOCV result in a total of 72 samples.
^gLOOCV result in a total of 57 training samples.
^hIn [20], only 3 of 8 AML testing samples in the dataset were mentioned. Thus, their test set contained 10 rather than 15 samples.
ⁱExcept for C4.5, all the others are not rule-based classifiers.

ISSN: 1755-8794