Ant colony based semi-greedy algorithm for regression tree induction

Regression trees belong to a very important class of regression models which allows to split feature space into segments with building specialized local model for each of them and to achieve visualizable, easy interpretable and accurate piece-wise models. In this paper we propose a novel ant colony based semi-greedy algorithm for regression tree induction, combining techniques from both traditional regression tree induction algorithms and Ant Colony Optimization. The results of experiments on publicly available data sets show that the proposed algorithm outperforms conventional algorithms for regression tree induction in accuracy and results in less complex solutions.