摘要
针对基于密度峰值的聚类算法(CFSFDP)无法自行选择簇中心点的问题,提出了CFSFDP改进算法。该算法采用簇中心点自动选择策略,根据簇中心权值的变化趋势搜索"拐点",并以"拐点"之前的一组点作为各簇中心,这一策略有效避免了通过决策图判决簇中心的方法所带来的误差。仿真实验采用5类数据集,并与DBSCAN及CFSFDP算法进行了对比,结果表明,CFSFDP改进算法具有较高的准确度及较强的鲁棒性,适用于较低维度的数据的聚类分析。
A new density peaks based clustering method （CFSFDP） was introduced in the paper. For the problem that it is difficult to decide the cluster number with CFSFDP, an improved algorithm was presentea, wltn a cluster center autu-matic choosing strategy, the algorithm search for the ＂turning points＂ with the trends of cluster center weight＇s changing. Then we could regard a set of points whose weight is bigger than ＂turning points＂ as the cluster center. The error brought by ruling in the decision graph could be avoided with the strategy. Experiment was done to compare to DBSCAN and CFSFDP with 5 kinds of datasets. ＇The results show that the improved algorithm has better performance in accuracy and robustness, and can be applied in clustering analysis for low dimension data.
出处
《计算机科学》
CSCD
北大核心
2016年第7期255-258,280,共5页
Computer Science
