Research Of Hybrid Clustering Algorithm Based On Ant Colony Algorithm

In the cluster analysis, the complexity of practical application and multiformity of data always make an algorithm only can be applied to a specific type of data, and the algorithm always has some defects. So, many new clustering algorithm are introduced continually. The clustering algorithm based on the chemical recognition system of ants(AntClust) is a new simulating biological evolution algorithm. It can produce the number of clustering automatically, find clusters of arbitrary shape, and process data of arbitrary type. It also has strong robustness and suitability. However, AntClust is based on random selection, and uses many statistic random parameters as judgement standard. These tow methods make AntClust have more computational mistake. Simultaneously, its method of deleting small nets is too rigid. This paper introduce a new hybrid clustering algorithm by means of improving AntClust and combining AntClust and K-means algorithm. This new algorithm can raise the quality of clustering result.According to the defects of AntClust in the initialization of ants, behavioral rules, computation of similarity and deleting small nets, this paper introduce new rules of random meeting, new method of computing similarity, new definition of complex clustering center and new rules of deleting small nets. In the improvement process, this paper uses the thought of means in behavioral rules. This method reduces the error caused by random and increases the precision of computation. New complex clustering center make the algorithm can compute the mean value of category attribute, raises the handling capacity of the algorithm and lays the foundation for combining with K-means. The improved AntClust reduces the number of iterations, and raises the quality of clustering result. According to the trait of improved AntClust, this alters K-means algorithm properly: Set the maximum number of iterations, which can stop the algorithm easily; In the process of iterations, algorithm can update the acceptance threshold of ants. This paper uses improved AntClust as overall frame, put K-means algorithm into this frame, and processing data by invoking the two algorithms repeatedly. The hybrid algorithm has higher computational accuracy than AntClust.Experimental results no UCI data sets prove that the improvement of AntClust and the combination of improved AntClust and K-means are effective. The hybrid can find effective clusters. At the same time, it can raises the quality of clustering result.