A new evaluation measure Q-statistic that incorporates the stability of the selected feature subset

Abstract

Classification problems in high-dimensional data with a small number of observations became more common, especially in micro-study data. Over the past two decades, many basic classification models and feature selection algorithms (FS) have been proposed to increase the accuracy of predictions. However, the result of the FS algorithm based on predictive accuracy will be unstable on differences in the training package, especially in high-dimensional data. This paper proposes a new measurement of the Q-statistic that includes the persistence of the subset of the specific features as well as the accuracy of the prediction. Next, we propose an enhanced FS algorithm that enhances the Q value of the algorithm applied. Experimental studies based on synthetic data and 14 microarray data sets show that Booster not only enhances the Q statistic value but also the predictive accuracy of the applied algorithm unless the data set is intrinsically correct to predict the use of the given algorithm.