Mirnacle: machine learning with SMOTE and random forest for improving selectivity in pre-miRNA ab initio prediction

Table 1 Comparing different combinations of methods for imbalanced data and learning algorithms. A 10-fold cross validation was performed in each case using TSHA1, TSHA2, and TSHA3 for the respective stages

Method	1st stage			2nd stage			3rd stage
	SN	SL	GM	SN	SL	GM	SN	SL	GM
LibSVM:
Cost matrix	4.7	64.7	17.4	3.8	81.8	17.7	4.2	100	20.6
Sampling	99.6	55.4	74.3	99.6	54.8	73.8	100	65	80.6
SMOTE	87.5	99.6	93.4	82.5	100	90.8	97.9	100	98.9
SMO:
Cost matrix	80.9	17.9	38	85.6	40	58.1	97.9	77.5	87.1
Sampling	84.3	77.7	81	86	90.6	88.3	98.7	96.7	97.7
SMOTE	86.3	83.5	84.9	91.9	94.4	93.1	99.9	99.3	99.6
MLP:
Cost matrix	74.1	16.7	35.2	79.2	49.7	62.7	98.7	3.9	19.6
Sampling	78.8	77.5	78.2	89.8	88.3	89.1	97	97.9	97.4
SMOTE	91.5	90.8	91.1	98	97.1	97.5	99.9	99.7	99.8
RF:
Cost matrix	46.2	44	45.1	74.6	76.5	75.5	89	89	89
Sampling	84.3	78.7	81.4	87.7	87.7	87.7	97.9	97.1	97.5
SMOTE	98.2	96.3	97.2	99.1	98.4	98.7	99.9	99.4	99.7

ISSN: 1471-2105