Abstract: The traditional DBSCAN algorithm needs to manually determine Eps and MinPts parameters, and the choice of parameters directly determines the rationality of the clustering results, thus this paper puts forward a new self-adaptive parameter determination method for DBSCAN algorithm. The method, based on the parameter optimization strategy, uses the data set’s own distribution characteristics to generate candidate Eps and MinPts parameters, automatically finds the cluster number change stable interval of the clustering result, and uses the Eps and MinPts parameters corresponding to the minimum density threshold in the interval as the optimal parameters. The experimental results show that the method can fully automate the clustering process and can select reasonable Eps and MinPts parameters, and the method obtains clustering results with high accuracy.

Key words: DBSCAN algorithm, self-adaptive, parameter optimization, K-average nearest neighbor

摘要： 传统DBSCAN算法需要人为确定[Eps]和[MinPts]参数，参数的选择直接决定了聚类结果的合理性，因此提出一种新的自适应确定DBSCAN算法参数算法，该算法基于参数寻优策略，通过利用数据集自身分布特性生成候选[Eps]和[MinPts]参数，自动寻找聚类结果的簇数变化稳定区间，并将该区间中密度阈值最少时所对应的[Eps]和[MinPts]参数作为最优参数。实验结果表明，该算法能够实现聚类过程的全自动化并且能够选择合理的[Eps]和[MinPts]参数，得到了高准确度聚类结果。

关键词: DBSCAN算法, 自适应, 参数寻优, K-平均最近邻法