Harmony-based feature weighting to improve the nearest neighbor classification

Abstract

This paper introduces the use of Harmony Search with novel fitness function in order to assign higher weights to informative features while noisy irrelevant features are given low weights. The fitness function is based on the Area Under the receiver operating characteristics Curve (AUC). The aim of this feature weighting is to improve the performance of the k-NN algorithm. Experimental results show that the proposed method can improve the classification performance of the k-NN algorithm in comparison with the other important method in realm of feature weighting such as Mutual Information, Genetic Algorithm, Tabu Search and chi-squared (χ2). Furthermore, on synthetic data sets, this method is able to allocate very low weight to the noisy irrelevant features which may be considered as the eliminated features from the data set. © 2013 Springer-Verlag.