This paper presents a novel ant-based clustering algorithm. Different from conventional ant-based clustering algorithms in which the objects are picked up or dropped down by virtual ants, our propose algorithm looks each object as an ant. The objects are first projected to a plane randomly. Then an attractive force field is created. The object (ant) is attracted by the similar ones in its local surrounding and moves to a certain place. The moving direction and moving range are determined by the composite of all attractive forces. The clusters are created by this attractiveness after many iterative cycles. The paper gives the detailed process of the algorithm. The performance of the algorithm is compared with other classic algorithms on several simulated and real datasets. The results are very encouraging in terms of the computation efficiency and clustering quality.
展开▼
