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.
CITATION STYLE
Agre, G., & Dzhondzhorov, A. (2016). A weighted feature selection method for instance-based classification. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9883 LNAI, pp. 14–25). Springer Verlag. https://doi.org/10.1007/978-3-319-44748-3_2
Mendeley helps you to discover research relevant for your work.