GA-SVM wrapper for feature selection

This paper discusses a wrapper approach for feature selection. Feature selection is performed by generating random collections of features, and then refining these features using genetic algorithm (GA). A support vector machine (SVM) is used as a classifier for different feature subsets during the feature selection stage. To indicate the best accuracy/complexity trade-off, a combination function with 5-fold cross-validation accuracy and the number of features are used to derive fitness scores. Comprehensive experiments on two data sets in the UCI machine learning repository confirm the effectiveness of the proposed strategy.