Tree Model Induction from Texts in Discharge Summaries

Abstract

This paper investigates the characteristics of decision trees for discharge summaries classifier, focusing on (1) selection of features in text mining and (2) splitting measures. The results shows that decision tree induction methods select attributes independently of the importance obtained by morphological analysis, which have small changes in the choice of splitting meatures. Thus, decision tree methods are found to be the simplest and powerful methods for mining classification knowledge from discharge summaries.