Evolutionary Algorithms for Global Decision Tree Induction

Abstract

Decision trees represent one of the main predictive techniques in knowledge discovery. This chapter describes evolutionary-induced trees, which are emerging alternatives to the greedy top-down solutions. Most typical tree-based systems search only for locally optimal decisions at each node and do not guarantee the optimal solution. Application of evolutionary algorithms to the problem of decision tree induction allows searching for the structure of the tree, tests in internal nodes, and regression functions in the leaves (for model trees) at the same time. As a result, such globally induced decision trees are able to avoid local optima and usually lead to better prediction than the greedy counterparts.