An Improved Weighted Feature Abstracting Algorithm

Abstract

In non-supervised data set, the importance of each feature is different. If the feature is setted with a proper weight, which can fully considers the lever of the influence on the cluster effect, then the clustering result will be improved. A feature evaluate function is proposed to obtain a set of feature weight vectors by minimizing the function, which is a multi-objective problem. So a fast and elitist multi-objective genetic algorithm is used to solve the problem and obtain the weight of feature. Finally, the weight of feature is introduced into the standard K-Means algorithm and the experiments on the UCI dataset show the validity of the algorithm.