A search based view of decision trees induction
Rafał Biedrzycki, Jarosław Arabas · 2010
Abstract. This paper addresses the issue of the decision tree induction. We treat this task as a search problem in the space of decision trees. We show that the search space definition allows for easy application of general search methods to solve this task. On the other hand it is possible to interpret standard problem-specific algorithms (e.g. ID3 and C4.5) as instances of specific search methods. We assume a certain metric in the space of decision trees, and define an evolutionary algorithm and a Monte Carlo method to search that space. We provide the experimental comparison of the aforementioned algorithms with ID3 and C4.5.