Selecting Salient Features in High Feature to Exemplar Ratio Conditions

This thesis research offers two contributions: (1) a new user-friendly graphical user interface that can be used in a feature screening process, (2) a new algorithm for feature screening in a situation where there is a high ratio of feature to exemplars. The first objective is achieved by creating a MATLAB based graphical user interface, which is named as STNGER. STNGER is evaluated on both abstract and the real life problems and provides promising results. For the second objective a new algorithm is suggested. This new algorithm is based on the SNR screening method. By means of this new algorithm, the SNR screening method can determine the salient features in a situation where there is a high ratio of features to exemplars. The performance of the new algorithm suggests that one can apply the SNR screening method to randomly chosen subsets of the data and retain the best features from each subset for subsequent analysis. In this fashion, noisy features are removed while creating new subsets with salient features. The new algorithm is demonstrated on both real-life and the well-defined abstract problems.