Acuna, Edgar (2003), A Comparison of Filter and Wrapper Methods for Feature Selection in Supervised Classification, Computing Science and Statistics, 35, I2003Proceedings/AcunaEdgar/AcunaEdgar.paper.pdf
In this paper we carry out an empirical comparison of the performance of filter and wrapper procedures for feature selection in supervised classification. The filter methods considered are the RELIEF, Las Vegas Filter, and a new procedure that is being introduced here called FINCO. Among the wrapper methods we considered sequential forward selection, sequential backward selection and the sequential floating forward selection. The classifier used for the wrapper methods is one based on kernel density estimation. Both type of procedures are compared according to their percentages of features selected and their effect in the misclassification error rate of a kernel density estimate classifier. The comparison is carried out in twelve datasets coming from the Machine Learning Database Repository at the University of California, Irvine.