Based On Swarm Intelligence, Ant Colony Clustering Algorithm And Applications

Swarm intelligence comes of the scientists' research and observation on the social insect. The so-called swarm intelligence is that a great many of simple, unintelligent units unite into a group and express intellectual behaviors through mutual cooperating with each other. Swarm intelligence exhibits a number of interesting properties such as flexibility, robustness, decentralization and self-organization. It has widely used in portfolio optimization problem, knowledge discovery, communication networks, data-mining and etc. The inspiration of ant colony clustering comes of the accumulation of ant bodies and classification of ant larvae. The classic ant colony clustering algorithm takes use of the characteristics of positive feedback of the ant colony. Such algorithm is robust, good convergence, and parallel. However, it is also with the disadvantage of long time convergence, easy stagnation and local optimization.The article proposed a new ant colony optimization clustering algorithm. This algorithm is based on a classics algorithm - LF algorithm. By the introduction of a new formula and the probability of similarity metric conversion function, as well as the new formula of distance, this algorithm can deal with the category data easily. Based on the traditional ant algorithm, it also introduces a new adjustment process, which adjusts the cluster generated by the carry process iteratively. We approve that that the algorithm can improve the efficiency and the convergence of the clustering theoretically. Data experiments show that the Optimized Ant Colony Algorithm can form more accurate and stability clusters than the K-Modes algorithm, Information Entropy-Based Clustering Algorithm, LF Algorithm. Scalability experiments show that the algorithm running time has an obvious linear relationship with the size of data set. Furthermore, we describe the process and idea of the algorithm usage by a mobile customer classification case and analyze the clustering results.In conclusion, compared with the traditional clustering algorithm, the Optimized Ant Colony Algorithm can handle large category dataset more rapidly, accurately and effectively, and keep the good scalability at the same time.