Evaluation of data discretization methods to derive platform independent isoform expression signatures for multi-class tumor subtyping

Table 4 Comparison of classification accuracy using top 100 features for platform transition.

Feature selection	CV (%)				SVM-RFE (%)				RF_based_FS (%)
Classifier	FC	Equal-W	Equal-F	k-means	FC	Equal-W	Equal-F	k-means	FC	Equal-W	Equal-F	k-means
SVM	40.8	26.3	84.2	81.6	36.8	40.8	85.5	39.5	28.9	30.2	76.3	39.5
RF	67.1	73.7	89.5	76.3	55.2	60.5	86.8	80.2	56.6	81.6	90.8	85.5
NB	25.0	23.7	80.2	71.0	32.9	23.7	76.3	22.3	23.7	23.7	84.2	36.8
PAM	35.5	23.7	78.9	64.5	39.5	27.6	73.7	32.9	39.5	23.7	81.6	44.7

The classification models were trained on exon-array (342 samples) and tested on RNA-seq (76 samples) data. Highest accuracy for each classification method is marked in bold. Only RF with Equal-F binning achieved greater than 90% classification accuracy.

ISSN: 1471-2164