Design And Implementation Of Cluster System Based On Data Field

With the rapid development of Internet technology, it brought great convenience for people’s daily life and work. It also generated lots of electronic data. The sizes and types of these data are vary. It is difficult to sorting them in accordance with the existing experience. And clustering analysis is useful for extracting valuable information.An hierarchical clustering method based on data fields can find the cluster of any shape. And it can effectively deal with noise points. The algorithm itself does not need to enter any parameters. It has quite good practical value. However, this clustering algorithm process is complex. And it can’t complete the task of cluster analysis based on the path cluster data sets and high-dimensional data sets. So it is necessary to improve the algorithm.Based on the above, on the basis of An hierarchical clustering method based on data fields, this paper proposes a new clustering algorithm based on potential field data. It successfully solves former’s defect.This paper’s the main content is as follows:1. On the basis of potential field data, this paper proposes a new clustering method. Creatively combining the potential difference and distance between data points, this paper finds a good and easier way to detect cluster. In the nearest higher point as the clustering direction, the algorithm links entire data set together. Then the whole data set is divided into several clusters according to their distribution characteristics.2. This paper used four data sets which commonly used in clustering algorithm to verify the ability to detect the cluster with complex shape. Then the proposed algorithm was contrast to two classical clustering algorithm clustering Kmeans and Dbscan. And it is proved that the proposed clustering algorithm was not less than the above two algorithms in terms of the quality and the efficiency of the clustering.3. In this paper, the processing method of noise points is studied. Based on data potential field, an effective method is proposed to identify the noise points.4. In this paper, a human face database was processed on the proposed clustering algorithm. And it was compared with peak density clustering algorithm which was published on Science. It was demonstrated the ability of this article clustering algorithm to identify different face on the data set is stronger than the latter.