Using Early Stopping to Reduce Overfitting in Wrapper-Based Feature Weighting

Abstract

It is acknowledged that overfitting can occur in feature selection using the wrapper method when there is a limited amount of training data available. It has also been shown that the severity of overfitting is related to the intensity of the search algorithm used during this process. We demonstrate that the problem of overfitting in feature weighting can be exacerbated if the feature weighting is fine grained. With greater representational power we risk learning not only the signal, but also the idiosyncrasies of the training data. In this paper we show that both of these effects can be ameliorated by the early-stopping strategy we present. Using this strategy feature weighting will outperform feature selection in most cases.