对DBSCAN聚类算法的改进.pdf

计算机研究与发展 ISSN1000．1239／CN11-1777／TP Journalof Researchand 42(增刊)：50～54，2005 Computer Development 对DBSCAN聚类算法的改进 袁 方1’2 孟增辉2 于 戈1 1(东北大学信息科学与工程学院沈阳110004) 2(河北大学数学与计算机学院保定071002) (yuanfang@mail．hbu．edu．en) An DBSCAN Improved ClusteringAlgorithm Yuan YuGel Fan91，-，MengZenghui2，and and 1 Science 10004) 1(CollegeofInformation Engineering，NortheasternUniversity，Shenyang Mathematwsand Science，Hebei 2(Collegeof Computer University，Baoding，071002) from withnoise．But AbstractDBSCAN candealwith cluster datasets algorithm arbitraryshaped spatial the constrainsits inthis isanew the forusersdetermine way parameters performance．Presentedpaper methodtodeterminethe MinPtsand fortheDBSCAN themethodisused parameters Eps algorithm．First is tolooksfor todeterminethesuitableMinPtsvaluebasedonthecurvesof themethodused Eps．Second has obvious inMinPts—thcurveof andtoconsiderthe ofthis asthe the that an Eps point point change Eps valueofthe showthatthe hasanautomaticof Eps．Experimentsimprovedalgorithm process parameter the DBSCAN andhasobtained MinPtsand with determiningparameters Epscompared original algorithm better performance． words basedon Key clustering；clusteringdensity；DBSCANalgorithm