Significance of classification scores subsequent to feature selection

Feature selection prior to classification is shown to inflate performance. Empirical distributions are used to estimate statistical significance of classification scores. In an example study, eleven high classification scores are obtained but only three are found to be significant at p = 0.05.