Research On Ant Colony Clustering Algorithm Based On Swarm Intelligence

With the rapid development of network technology, gradually increasing the amount of data accumulated by the various trades, how to dig out the useful information form the vast amounts of data become more and more the attention of researchers. Data mining technology has been used as a development opportunity by many companies. Data mining is extracted from the mass of data implied, position, and the potential value of the knowledge and rules. Classical data mining algorithms have cluster analysis, Bayesian classification, association rules, support vector machines, genetic algorithms, among cluster analysis is an important work in the mining, the main role is that the data of the same category gather together in the data and dissimilar data are separated from each other. The most popular algorithms are k-means and k-medoids algorithms in cluster analysis. Although the idea of two cluster analysis algorithm is very simple and also widely used, there are some shortcomings. For example, the initialization grouping largely determines the clustering results, and you enter the specified value of k, in addition, the order of the input data is also very sensitive. When the large amount of data, the number of iterations will increase and efficiency will be reduced and so on.So we want to make clustering analysis more "intelligent", and clustering algorithm is not sensitive to the order of the input data or no input priori knowledge of the value of k, it forms clustering. In recent years, with the generation of swarm intelligence algorithm, swarm intelligence is applied to clustering algorithm by number of researchers, and achieves good experimental results.Swarm intelligence algorithm mainly includes particle swarm algorithm and ant colony algorithm, they simulate natural birds and ant colony. Because swarm intelligence algorithms simulate the natural social animals, there are also some shortcomings. The ant colony algorithm needs to set many parameters. In ant colony clustering algorithm, ants need to do a large number of exploratory random motions before the data object is picked up or put down, so it takes a very long time and has a less efficient. Moreover, the algorithm has the strong sensitivity of input parameters. This article aims at the shortcomings of ant clustering algorithm to propose the improved algorithm, the major improvements are as follows:Firstly, algorithm improves the method of similarity measure, the different attributes of each data are not equivalent in the treatment. In other words, each attribute is different contribution for the data belongs to which class. Improved metric increases the distance between the discrimination properties. Thus obtained similarity will be more meaningful than the conventional similarity measure.Secondly, this paper proposes a new pick-up strategy. Ants will pick up the data group which has been formed micro-clusters after iterate several times. This can greatly improve the efficiency of the algorithm. In addition still in the case of failure putting down load objects, algorithm adds strategy that ants pick up the similar data object. The asocial data objects are removed.Finally, this article increases the improved ant memory and noise data processing. The memory strategy adopts in the paper has the required a similarity threshold. Only when similarity between data objects exceeds this threshold, ants move to the position. If the number that ants loaded data objects putting down failure numbers exceeds a given threshold, ants will place it in an empty grid.This paper presents an improved ant colony clustering algorithm after experimental datasets Iris validation. Efficiency and accuracy are superior to the traditional ant colony clustering algorithm.