Feature subset selection for rule induction using RIPPER
Jihoon Yang, Asok Tiyyagura, Fajun Chen, Vasant Honavar · 1999
The choice of features or attributes used to represent patterns in the synthesis of pattern classifiers using machine learning algorithms has a strong impact on the accuracy of the classifier, the number of examples needed to attain a given classification accuracy on test data, the cost of classification, and the comprehensibility of the learned classifier. This presents us with a feature subset selection problem, namely, the selection of a subset of features from a much larger candidate set of features to represent patterns to be classified so as to optimize multiple criteria such as the accuracy and the cost of pattern classification. Evolutionary algorithms, because of their ability to find good solutions offer a promising approach to such a multicriteria optimization problem. Results of experiments reported in this paper demonstrate that feature subset selection using a genetic algorithm results in substantial improvement in classification accuracy and comprehensibility, and substa...