Robust Instance-Based Feature Selection for Density Estimation

Abstract

Classification issues in high dimensional information with alittle range of observations are getting additional common particularly in microarray information. throughout the last twenty years, voluminous economical classification models and have choice (FS) algorithms are projected for higher prediction accuracies. However, the results of associate degree FS algorithmic program supported the prediction accuracy are unstable over the variations within the coaching set, particularly in high dimensional information. This paper proposes a replacement analysis live Q-statistic that includes the steadiness of the chosen feature set additionally to the prediction accuracy. Then, we tend to propose the Booster of associate degree FS algorithmic program that enhances the worth of the Q-statistic of the algorithmic program applied. Empirical studies supported artificial information and fourteen microarray information sets show that Booster boosts not solely the worth of the Q-statistic however additionally the prediction accuracy of the algorithmic program applied unless the information set is per se troublesome to predict with the given algorithmic program.