Evolutionary decision tree induction with multi-interval discretization

Decision trees are one of the widely used machine learning tools with their most important advantage being their comprehensible structure. Many classic algorithms (usually greedy top-down ones) have been developed for constructing decision trees, while in recent years evolutionary algorithms have found their application in this area. Discretization is a technique which enables algorithms like decision trees to deal with continuous attributes as well as discrete attributes. We present an algorithm that combines the process of multi-interval discretization with tree induction, and introduce especially designed genetic programming operators for this task. We compared our algorithm with a classic one, namely C4.5. The comparison results suggest that our method is capable of producing smaller trees.