Skip to main content

Table 2 Weighted area under ROC curve for different feature selection and classifier induction methods showed J48 with significance-based feature selection as the one of the best performing and immune to overfitting modeling approaches. Significance filter was chosen to be preferable based on the highest average across multiple feature selection methods. J48 has been chosen as the best method based on the highest average across all methods

From: Expression-based decision tree model reveals distinct microRNA expression pattern in pediatric neuronal and mixed neuronal-glial tumors

Dataset J48 ZeroR Jrip Decision Stump Random Tree SVM Average
Full dataset 0,59 0,50 0,63 0,66 0,59 0,61 0,60
Significant miRNAs 0,71 0,50 0,71 0,60 0,67 0,50 0,62
Support vector machine-based attribute evaluator 0,66 0,50 0,62 0,49 0,61 0,50 0,56
TTP 0,55 0,50 0,52 0,49 0,53 0,51 0,52
Average 0,63 0,50 0,62 0,56 0,60 0,53  
  1. SVM Support vector machine with radial basis function kernel, TTP Targeted projection pursuit