Improving the performance of evolutionary algorithms for decision rule learning

Evolutionary algorithms appear as an interesting alternative to achieve minimal error rates and low numbers of rules in supervised learning tasks. In spite of the computational cost of this approach, some proposals can be applied to make the algorithm faster and more efficient. This paper describes some of these proposals, which are integrated in the evolutionary tool HIDER∗. Specifically, we developed a new genetic encoding for the individuals of the evolutionary population and a novel data structure for the evaluation process. These approaches allow the evolutionary algorithms to reduce the high computational cost and to obtain high quality solutions.