Table 3.

Summary of the SVM models optimized against different criteria, including the optimized SVM parameters, correct rates in a 7-fold CV, and predictive performances for CYP2D6.

	Optimization Criteria
	accuracy	precision	recall	AUC
C	2.5	0.5	64.0	4.0
γ	1.0	0.5	0.3	0.5
Rate	87.4%	93.0%	52.5%	86.0%
AUC-ROC	0.76	0.76	0.7	0.76

Abbreviations: $accuracy=\frac{TP+TN}{P+N}$; AUC-ROC = area under the receiver operator characteristic curves; C = cost of misclassification; CV = cross-validation; γ = gamma in kernel function; $precision=\frac{TP}{TP+FP}$; $racall=\frac{TP}{P}$; P = positive; N = negative; TP = true positive; TN = true negative; FP = false positive.