Ant Colony Clustering Algorithm

As a new kind of intelligent optimization method, ant colony algorithm (ACA) has the features of distributed parallel calculation, information positive feedback and heuristic search ability, and has demonstrated its outstanding performance and great potential for development in solving many complex optimization problems. Applied ACA into the area of clustering analysis, has become an intelligent way to deal with the phenomenon of "information exploding but knowledge poor" in the information age. As the ACA itself is still in the development stage, and requires a lot of verify work, so it is of great significance to conduct a comprehensive, in-depth study on the ant colony clustering algorithm (ACCA).This dissertation studied and analyzed ACCA deeply and proposed an improved algorithm. The main work is as follows.1. Summarized some representative ant colony clustering algorithms in recent years. First of all, gave a brief introduction and comparative analysis of two basic models of ACCA and typical algorithms based on them. Then, outlined the ideas of some improved ant colony clustering hybrid algorithms with representative.2.Proposed an improved ant colony clustering hybrid algorithm based on class-connectivity(IACCHA). IACCHA used the distributed search ability of ACA to avoid local optimum, and used the simplicity and high efficiency of K-means algorithm, the class-connectivity to enhance the performance efficiency.The improvement of the algorithm mainly contained the following aspects: set a threshold to reduce the isolation of pseudo-classes; adopted the nearest neighbor rule to amend the initial clustering results, and then conducted a second clustering to the cluster centers; before testing of the improved algorithm, used different data pre-processing technologies: information entropy method to determine attributes weights and principal component analysis to reduce dimensions; when testing algorithm, changed the interval of radius to test stability of the algorithm, and varied the step to get the best results of clustering.The IACCHA test of functionalities and performance reveal very encouraging results in terms of processing efficiency, clustering ability and stability, and indicate that the algorithm can be used to obtain global optimal solution.