Adaptive and fast density clustering algorithm

Abstract

As a density clustering method, DBSCAN clustering algorithm can automatically determine the number of clusters and effectively deal with the clusters of arbitrary shape, but the choice of global parameter Eps and MinPts require manual intervention and the region query process is complex and such query mode easily lose objects. In order to solve the above problems, improved adaptive parameters choice and fast region query density clustering algorithm. According to the KNN distribution and mathematical statistical analysis adaptively calculate the optimal global parameter Eps and MinPts, which avoids manual intervention and achieves full automation of the clustering process. Utilize the improved method to select the representative seed to operate region query, without losing objects, improved the efficiency of clustering. Experiment results at four typical data sets show that the proposed method effectively solves the difficulties of DBSCAN in parameter selection and efficiency.