适合大规模数据集的增量式模糊聚类算法

Incremental fuzzy (c+p)-means clustering for large data

FCPM算法已被成功地应用到模糊系统建模上,但其在某一类的聚类中心已知的大规模数据上的聚类性能较差。为了避免这个缺点,参照单程模糊c均值(SPFCM)聚类算法、在线模糊c均值(OFCM)聚类算法,提出了适合大规模数据集的增量式模糊聚类算法(Incremental fuzzy(c+p)-means clustering,IFCM(c+p))。通过在每个数据块中使用FCPM算法进行聚类,把每个数据块的聚类中心及其附近的一些样本点加入到下一个数据块参与聚类,同时添加平衡因子以提高算法聚类性能。同SPFCM、OFCM以及rse FCM算法相比,IFCM(c+p)对初始聚类中心不敏感。实验表明在没有花费很多运行时间的情况下,IFCM(c+p)算法的聚类性能比SPFCM算法和rse FCM算法更具优势,因此该算法更适合处理某一类聚类中心已知的大规模数据集。

FCPM has been demonstrated to be successful in fuzzy system modeling, however, it will be ineffective for large data clustering tasks where the cluster centers of one class are known. In order to circumvent this draw- back, referring to single-pass fuzzy c-means （SPFCM） clustering algorithm and online fuzzy c-means （OFCM） clustering algorithm, the incremental fuzzy clustering algorithm for large data called IFCM （c＋p） is proposed in this paper. FCPM algorithm is used to cluster for each data block at first, and then the clustering centers of data block and some of the sample points being near them are joined into the next block to be clustered, meanwhile the bal- ance factor is given to enhance the clustering performance. In contrast to SPFCM, OFCM and rseFCM, IFCM（c＋ p） is not sensitive to the initial cluster centers. The experiments indicate the proposed clustering algorithm IFCM （c ＋p） is competitive to the clustering algorithms SPFCM and rseFCM in the clustering performance without the loss of running time a lot, hence it is especially suitable for large data clustering tasks where the cluster centers of one class are known.