An Improved Clustering Algorithm Based on Ant-Tree

In this paper, we propose an improved clustering algorithm based on the Ant-Tree algorithm. This method represents a more flexible version of its basis. The classes with high density are defined as definite classes, and our algorithm starts with finding the definite classes. Centroid approximation method is utilized to make the clustering model of Ant-Tree more accurately by approaching the real center of the classes gradually. The ants that have fixed themselves on the structure can be disconnected from the tree for a better position, and in this way more accurate results of clustering can be achieved. As a consequence, this algorithm builds adaptively a tree structure which changes over the run in order to improve the final results. Compared against some other ant-based clustering algorithms, our approach acquires better results on some standard databases efficiently as demonstrated in experiments.