Table 2.

The performance ${}^1$ of our method by using different features.

FEATURE	AUC	ACC	MCC	SN	SP
k-Gram	0.7143	0.7312	0.3288	0.3532	0.9128
MMI	0.6750	0.7061	0.2430	0.2529	0.9237
DWT	0.8063	0.7593	0.4213	0.5057	0.8810
PseAAC	0.6997	0.7214	0.2936	0.2961	0.9256
Combination ${}^2$	0.8338	0.7725	0.4589	0.5540	0.8774
Combination (FS) ${}^3$	0.8632	0.8017	0.5558	0.7268	0.8377

${}^1$ The values were calculated using the testing results on benchmark dataset. The classifier was support vector machine (SVM), and the validation method was target-jackknife cross-validation. ${}^2$ Feature size was 612-D, including all of k-gram, MMI, DWT, and PseAAC. ${}^3$ Feature size was 114-D feature, selected by feature selection in SVM. AUC: area under the receiver operating characteristic curve; ACC: accuracy; MCC: Matthews correlation coefficient; SN: sensitivity; SP: specificity; FS: feature selection. The bold digits are the greatest values in each column.