A weighted feature selection method for instance-based classification

Abstract

The paper presents a new method for selecting features that is suited for the instance-based classification. The selection is based on the ReliefF estimation of the quality of features in the orthogonal feature space obtained after PCA transformation, as well as on the interpretation of these weights as values proportional to the amount of explained concept changes. The user sets a threshold defining what percent of the whole concept variability the selected features should explain and only the first “stronger” features, which combine weights together exceed this threshold, are selected. During the classification phase the selected features are used along with their weights. The experiment results on 12 benchmark databases have shown the advantages of the proposed method in comparison with traditional ReliefF.