Designing smaller decision trees using multiple objective optimization based GPs
Shinichiro Haruyama, Qiangfu Zhao · 2003
Decision tree (DT) is a good model for machine learning. Many methods have been proposed in the literature for designing DTs from training data. Most existing methods, however, are single-path search algorithms which provide only one of the possible solutions. In our research, we have tried to design DTs using the genetic programming (GP). Theoretically speaking, GP can generate many different DTs, and thus might be able to design smaller DTs with the same performance. In practice, however, the DTs obtained by GP are usually very large and complex. To solve this problem, we have proposed several methods for evolving smaller DTs. In this paper, we examine several multiple objective optimization (MOO) based GPs, and verify their effectiveness through experiments.