Abstract: In order to improve the training efficiency, an advanced Fuzzy Support Vector Machine(FSVM) algorithm based on the density clustering(DBSCAN) is proposed. The original data are pretreated through DBSCAN algorithm which is utilized to remove the center sample points that make little contribution to the classification. The remaining sets of edge sample complete the training work for FSVM to seek optimal hyper-plane. Experimental result shows that it can shorten the training time and improve the work efficiency while keeping the distribution of original sample and guaranteeing the classification precision.

Key words: Fuzzy Support Vector Machine(FSVM), density clustering, edge samples

摘要： 为了提高模糊支持向量机在数据集上的训练效率，提出一种改进的基于密度聚类(DBSCAN)的模糊支持向量机算法。运用DBSCAN算法对原始数据进行预处理，去除对分类贡献小的中心样本，用剩余的边缘样本集合完成模糊支持向量机的训练工作。实验表明，该方法形成的聚类边缘样本较好地保持了原样本的分布情况，在保证分类精度的同时，大大缩短了训练时间，提高了工作效率。

关键词: 模糊支持向量机, 密度聚类, 边缘样本

CLC Number: 

• TP18