Optimizing Accuracy and Size of Decision Trees

This paper presents the problem of finding parameter settings of algorithms for building decision trees that yield optimal trees—accurate and small. The problem is tackled using DEMO algorithm, an evolutionary algorithm for multiobjective optimization that uses differential evolution to explore the decision space. The results of the experiments on six datasets show that DEMO is capable of efficiently solving this problem, offering the users a wide choice of near-optimal decision trees with different accuracies and sizes in a reasonable time.