A dynamic clustering based on genetic algorithm

Abstract

The paper presents a dynamic clustering method based on genetic algorithm. In order to obtain the perfect clustering results, the preprocessing such as primary component analysis or wavelet transformation is often used, but it is likely to result in distortions. In this paper, the essential associations between objects are modeled by their dissimilarity. The dissimilarity between objects is mapped into their Euclidean distance, and then the mapping is optimized by genetic algorithm, which means the coordinates of each object are optimized by genetic algorithm gradually, and thus makes the Euclidean distances among objects approximate to their dissimilarity. The primary advantages of the proposed method are that the clustering does not depend on the feature space distribution of the input objects while simplifying the clustering and improving the visualization. A numerical simulation illustrates its feasibility and availability.