Comparative Analysis of Decision Tree Algorithm for Learning Ordinal Data Expressed as Pairwise Comparisons

Abstract

Decision Tree is a very mature machine learning method used to solve classification problems. In this paper, we show the review of Decision Tree implementation for learning user preferences data expressed in pairwise comparisons. Decision Tree can be considered as one of the suitable methods for this problem due to its white-box approach, so that we can evaluate the result and re-use the model for further analysis, such as giving a recommendation. We used 10-fold cross-validation and hold-out technique to evaluate the performance of four different decision tree algorithms. The result shows that some decision tree algorithms like J48 outperform the others for learning pairwise preferences on a specific training split point. This paper has demonstrated, through use cases and experiments of pairwise preference problem, the effectiveness of decision tree method, and of its novel use of learning ordinal data.