Research On Strategies For Improving Ant Colony Clustering Algorithm

Ant colony clustering algorithm is a clustering algorithm which is derived from nature bionics, with its unique advantages and strong vitality, causes wide attention of scholars at home and abroad. Ant colony clustering algorithm has a lot of characteristics such as parallel, self-organization, flexibility and robustness and so on. It can be applied to many fields with slide modification of the algorithm. Compared with the traditional clustering algorithm, the ant colony clustering algorithm has many advantages, but there is still a lot of room for improvement, such as the clustering efficiency and quality.Based on the study and research on ant clustering algorithm and a series of its improved algorithm, the paper analyzes and discusses the shortcomings of the algorithm from two different angles, then designs and implements two different improved ant colony clustering algorithm.First, the traditional ant colony clustering algorithm treats every feature of data equally, but in practice, the contribution rate of attributes is different from each other. Therefore, giving all features the same weight will eventually affect the quality of clustering. To overcome the defect, the method of principal components analysis is introduced into the ant colony clustering algorithm to calculate the contribution rates of attributes and to construct the weights of attributes.On this basis, combined with a new initialization strategy, an improved ant colony algorithm with weighted attributes is proposed in this paper.The results show that reasonable weight distribution can effectively improve the quality of clustering.Second, in essence the ant colony clustering algorithm is a method based on the distance metric.The same with the other clustering method based on distance measure, ant colony clustering is good at dealing with the ellipsoid distribution or Gaussian distribution data sets, but as for the boundary between the classes of the data sets is non-linear, the quality of clustering is often not unsatisfactory. To overcome the defect, the kernel methods is introduced into ant colony clustering algorithm, in which the data set in low-dimensional is mapped to high-dimensional Hilbert space by some non-linear means and accomplishes the clustering task in high-dimensional feature space,thereby to solve some linear inseparable problems in low-dimensional space.On this basis, combined with a new initialization strategy, an improved ant colony algorithm based on Gaussian-kernel is proposed in this pape.The result show that the Gaussian-kernel based ant colony clustering algorithm has a obvious effect on the improvement of efficiency and quality of clustering.Finally, this paper summarizes the work of this paper, and further research directions are discussed.