ROC Analysis of a Linguistic Decision Tree Merging Algorithm

ROC analysis is an important tool for evaluation and comparison of classifiers in imprecise environments (i.e., class distribution and cost parameters are unknown). Area Under the Curve of ROC (AUC) is increasingly being recognized as a better measure for evaluating algorithm performance than accuracy. A bigger AUC value implies a better ranking performance for a classifier. Linguistic decision tree (LDT) is a model based on a random set framework of modelling with words. Classification is made based on probability estimates from all leaves of the tree. In this paper, a branch merging algorithm for LDT model is proposed to generate more compact trees and no significant reduction in AUC values is found.