Medium and high-dimensionality attribute selection in Bayes-type classifiers

摘要

This paper introduces the application of attribute selection methods along with Bayes classifiers. The proposal has been evaluated in eleven binary and multi-class real data sets with a number of instances lower than a thousand and a number of attributes between eight and sixteen thousand. Among them, five data sets belong to the Bioinformatics area. Experiments show that, in general terms, the most convenient attribute selector is based on a correlation measure. According to the reported results, the best classifier in high-dimensional data sets is Naive Bayes.