Feature weighting by RELIEF based on local hyperplane approximation

Abstract

In this paper, we propose a new feature weighting algorithm through the classical RELIEF framework. The key idea is to estimate the feature weights through local approximation rather than global measurement, as used in previous methods. The weights obtained by our method are more robust to degradation of noisy features, even when the number of dimensions is huge. To demonstrate the performance of our method, we conduct experiments on classification by combining hyperplane KNN model (HKNN) and the proposed feature weight scheme. Empirical study on both synthetic and real-world data sets demonstrate the superior performance of the feature selection for supervised learning, and the effectiveness of our algorithm. © 2012 Springer-Verlag.