A Two-Step Feature Selection Method to Predict Cancerlectins by Multiview Features and Synthetic Minority Oversampling Technique

Table 5

Performance comparisons with the existing methods using 5-fold cross-validation.

Method	Sensitivity	Specificity	Accuracy	MCC	Feature number
Amino Acid Composition [20]	0.680	0.642	0.658	0.32	20
Dipeptide Composition [20]	0.673	0.628	0.648	0.30	400
Split based Composition (2-part) [20]	0.663	0.642	0.651	0.31	40
Split based Composition (4-part) [20]	0.651	0.669	0.661	0.32	80
Position-Specific Scoring Matrix [20]	0.679	0.686	0.683	0.36	400
PSSM with 14 PROSITE domains [20]	0.680	0.699	0.691	0.38	414
-gap dipeptides [22]	0.691	0.801	0.752	0.495	68
Our method	0.779	0.717	0.748	0.497	13