Research On Selection Of Clustering Algorithms Be Aimed At Ring And None-ring Clustering Structure

Clustering analysis is an important study field in data mining and application. Many clustering algorithms have been proposed so far. But different algorithms tend to give different results, and we still don’t have a best clustering algorithm to suit all possible data sets. It’s necessary to choose a suitable clustering algorithm for the data set to ensure the quality of result. So, some study on the selection of clustering algorithms has begun. Compared with choose an clustering algorithm for the data set blindly, selection of clustering algorithms can choose an appropriate clustering algorithm and get higher quality results.The paper first compare the results of several alternative clustering algorithms on different real data and artificial data. Then, as the basis for the algorithm selection,it analyzes the applicability of the algorithms to the data set with different cluster structure(For example, the data set contains ring cluster or the data set dosen’t contain ring cluster). We proposed a method named SCGM aimed at the selection of clustering algorithms. Firstly, the method divide data space into grid space and allocation data to grid space. Then, dual principle puts used to analyzes the relationship of cells of grid which hasn’t assigned data. To find whether the data contains a ring-shaped clustering by the number of Grid-MST. Finally, based on clustering structure to select an appropriately clustering algorithm. Proposed another method named ACS A to ensure the result of the clustering algorithm which was selected. Firstly, the method constructs the Grid-MST to analyzes whether the data set (D) has ring-shaped structure of clustering. Then, it selects an appropriate set of alternative clustering algorithms(Include k kinds of algorithms) for the data set(D) by the result. To find the optimal algorithm from set of alternative clustering algorithm, it uses the cluster validity index to assess the clustering results of k kinds of clustering algorithms which clustering the data set separately. Then, select a clustering results which has the optimal clustering evaluation values as the final result. The experiment results show that the proposed method can select the clustering algorithm successfully, and get high quality clustering results.