Kernel Function Clustering Based on Ant Colony Algorithm

Abstract

Cluster analysis is one of the several important tools in modern data analysis, and the clustering can be regarded as an optimization problem. The underlying assumption is that there are natural tendencies of cluster or group structure in the data and the goal is to be able to uncover this structure. In general, traditional clustering algorithms are suitable to implement clustering only if the feature differences of data are large. If the feature differences are small and even cross in the original space, it is difficult for traditional algorithms to cluster correctly. By using kernel functions, the data in the original space was mapped into a high-dimensional feature space, in which more feature of the data were exposed so that clustering could be performed efficiently. Ant colony algorithms are a novel category of evolutionary computing methods for optimization problems. Taking advantage of ant-based clustering algorithm and kernel method, we propose in this paper a kernel function clustering based on ant colony algorithm. The kernel method is extended to ant-based clustering algorithm. In contrast to existing method, AKC has better robustness and convergence. It not only can get better clustering effect at the circumstance of outliers existing, but also can make the final clustering results less sensitive to the number of beforehand clustering data because of relaxed subjection degree condition.