Table 3.

Comparison of the performance of the classification-based models with MDR

Feature	Method	RF	NB	kNN	MDR
					Two-way	Three-way
Validity criteria	Classification accuracy	0.734	0.702	0.733	0.647	0.721
	F-measure	0.853	0.785	0.841	0.764	0.861
	Precision	0.743	0.845	0.754	0.675	0.772
	Recall	0.998	0.734	0.954	0.664	0.883
Overfit		Very resistant since boot strap selection is performed	Relatively risky	Boot strapping performed to avoid overfit	Risky k-fold cross-validation used to overcome overfit problem
Advantages		Nonparametric Interpretable Resistant to noise	Resistant to noise Good for eliminating missing values	Simple, flexible Arbitrary decision boundaries	Nonparametric test Flexible Evaluate interactions
Disadvantages		Sensitive to inconsistent data	Accuracy degraded by correlated variables Nondeterministic	Sensitive to noise	Too slow High computation burden

Abbreviations: RF, random forest; NB, naïve Bayes; kNN, k-nearest neighbor; MDR, multifactor dimensionality reduction.