Title: ComboSplit: Combining Various Splitting Criteria for Building a Single Decision Tree

Year of Publication: Nov - 2014
Page Numbers: 1-8
Authors: Md Nasim Adnan , Md Zahidul Islam
Conference Name: The International Conference on Artificial Intelligence and Pattern Recognition (AIPR2014)
- Malaysia


Typically existing decision tree building algorithms use a single splitting criterion such as Gain Ratio and Gini Index. In this paper three existing splitting criteria are compared within the framework of the C4.5 decision tree building algorithm. We also propose a technique called ComboSplit for combining the existing splitting criteria to build a single decision tree. We experimentally evaluate the decision trees obtained by various existing splitting criteria and ComboSplit. Ten publicly available datasets are used in the experiments. Decision Trees obtained by ComboSplit generally have higher prediction accuracy than the trees obtained by the existing splitting criteria.