PHDcleav: a SVM based method for predicting human Dicer cleavage sites using sequence and secondary structure of miRNA precursors

Table 4 Performance of different WEKA methods for Dicer cleavage prediction and their comparison with SVM model.

Methods	Sn	Sp	Ac	Mc	AUC
SVM ^light ( *Model 1* )	84.32	88.46	86.40	0.730	0.922
Random Forest	80.54	90.81	85.68	0.717	0.921
Naïve Bayes	81.26	86.30	83.78	0.676	0.909
Simple CART	81.08	85.40	83.24	0.665	0.879
REP Tree	80.54	81.08	80.81	0.616	0.872
Random Tree	69.54	74.05	71.80	0.436	0.752

All models were used for the prediction of cleavage site at 5p arm by using 'extended binary pattern' feature of 14 nt window taken from 'quikfold' derived structure (structure of CP-5p). Models were developed on training dataset using non-redundant 5-fold cross validation techniques.
Sn: sensitivity, Sp: specificity, Ac: accuracy, Mc: Matthews correlation coefficient. AUC: area under curve.

ISSN: 1471-2105