Less biased measurement of feature selection benefits

Abstract

In feature selection, classification accuracy typically needs to be estimated in order to guide the search towards the useful subsets. It has earlier been shown [1] that such estimates should not be used directly to determine the optimal subset size, or the benefits due to choosing the optimal set. The reason is a phenomenon called overfilling, thanks to which these estimates tend to be biased. Previously, an ouler loop of cross-validalion has been suggested for fighting Ihis problem. However, ihis paper points oui that a straightforward implementation of such an approach still gives biased estimates for Ihe increase in accuracy that could be obtained by selecting the besl-performing subset In addition, two melhods are suggested that are able to circumvenl this problem and give virtually unbiased results without adding almost any computational overhead. © Springer-Verlag Berlin Heidelberg 2006.