基于无监督聚类的约简支撑向量机

为解决标准支撑向量机算法所面临的巨大的计算量问题,Lee和Mangasarian提出了约简支撑向量机算法;但他们选取的“支撑向量”是从训练样本里面任意选的,其分类结果受随机性影响比较大。该文利用简单的无监督聚类算法,在样本空间中选取了一些具有较强代表性的样本作为“支撑向量”,再运用约简支撑向量机算法,有效地减少了运算量。实验验证文中方法可以用较少的“支撑向量”来得到较高的识别率,同时运行时间也大大缩短。

Unsupervised Clustering Based Reduced Support Vector Machines

To overcome the heavy computation of the standard Support Vector Machines(SVMs ),Lee and Mangasarian have proposed Reduced Support Vector Machines(RSVM),but the method they used to select“support vectors”is to choose them from the training set randomly,and this will affect the test result.In this paper,some representative vectors are selected as support vectors via a simple unsupervised clustering algorithm,and then RSVM method is applied on these vectors.The experimental results demonstrate that compared with the standard RSVM method,the proposed method can get higher recognition accuracy with fewer“support vectors”and the running time is reduced significantly.