Improvement Of Clustering Algorithm And Its Application

The clustering algorithm is a hot topic in statistical research,and it is an important tool for data mining and large data.In this paper,the clustering algorithm is analyzed and the following results are obtained.Firstly,a clustering algorithm based on the perturbation factor-based criterion function is proposed.To solve the shortcomings of the K-means clustering algorithm which is sensitive to the initial cluster centers,an ant colony clustering algorithm is proposed to search K-means clustering algorithm,and a perturbation factor which obeys the uniform distribution is added to the clustering criterion function,and the clustering algorithm based on the perturbation factor criterion function is established.Then the ant colony clustering algorithm,the original clustering algorithm and the improved clustering algorithm are compared with experiments.The experimental results show that the improved clustering algorithm has better clustering effect than the other two and the clustering result is more stable.Secondly,a clustering algorithm based on similarity degree of perturbation factor is set up.K-means clustering algorithm are aimed at the shortcoming of easily trapped in local minimum,to add to the range of a set of search space narrowing of random sequence,based on disturbance factor under the similarity of the clustering algorithm.Then the contrast experiments of the original and the improved clustering algorithm are made.The experimental results show that the improved clustering algorithm has better clustering ability and higher accuracy than the before.Thirdly,the ant colony clustering algorithm based on weighted Mahalanobis distance is set up.As to the original algorithm which is neglected the influence of the correlation between the variables and the influence of the dimension,the weights are determined by the coefficient of variation method which is used to measure the important degree of the attribute.An improved ant colony clustering algorithm based on weighted Mahalanobis distance is proposed and compared with the FCM,M-FCM and ACCA,and the results show that the improved clustering algorithm has higher clustering accuracy than the other two,then the M-ACCA is used to research the stocks' cluster and the results are compared with the ACCA.The results show that the clustering results of M-ACCA are better than that of ACCA.Finally,the contents and methods of this paper are summarized,and the futureresearch problems and directions are also discussed.