Data Mining applications generally use learning algorithms in order to induce knowledge. In domains where explanation about classification decisions is essential, symbolic supervised learning algorithms are appropriated. To scale up learning algorithms to deal with large databases, data sampling techniques can be applied. Afterwards, learning algorithms can be used on each sample to induce a set of classifiers which can be combined into an ensemble of classifiers or into a unique classifier. In this work we consider the latter approach and propose the use of a genetic algorithm. We have implemented the genetic algorithm and several evaluation functions into a computational environment for evolving sets of knowledge rules, described in this work as well as experiments carried out on several datasets. Good experimental results were obtained by the genetic algorithm.
|