关键词: DBSCAN算法, 数据场, 聚类

Abstract: DBSCAN (density based spatial clustering of applications with noise) algorithm is a typical density-based clustering algorithm. The algorithm can discover the arbitrary-shaped clusters. However, the clustering results depend on the two parameters Eps and MinPts which are chosen by users. And for some datasets with large density differences, either the clustering results may have the incorrect cluster number, or the algorithm may label part of the data as noise. Using the advantages that data field can commendably describe the data distribution and reflect the data relationship, this paper proposes a new clustering algorithm called improved DBSCAN algorithm based on data field. The algorithm introduces the concept of average potential difference and dynamically determines Eps and average potential difference of each class during the clustering process. In this way, it can receive better clustering results for some clusters with large density differences. Experimental results indicate that the proposed algorithm performs better than DBSCAN algorithm.

Key words: DBSCAN algorithm, data field, clustering