Abstract: Based on hierarchical clustering and re-sampling, this paper presented a Support Vector Machine (SVM) classification method for large-scale data, which combined supervised learning with unsupervised learning. The proposed method first used k-means cluster analytical technology to partition dataset into several subsets. Then, the method clustered class by class for each subset and selected samples in each clustering center neighborhood to form candidate training datasets. Last, the method applied SVM to train and model for candidate training datasets. The experimental results show that the proposed method can substantially reduce SVM learning cost. Meanwhile, the proposed method has better classification accuracy than random re-sampling method, and can attain about the same classification accuracy of the non-sampling method.

Key words: large-scale data, classification, clustering, re-sampling, Support Vector Machine (SVM)

摘要： 针对大规模数据的分类问题,将监督学习与无监督学习结合起来,提出了一种基于分层聚类和重采样技术的支持向量机(SVM)分类方法。该方法首先利用无监督学习算法中的k-means聚类分析技术将数据集划分成不同的子集,然后对各个子集进行逐类聚类,分别选出各类中心邻域内的样本点,构成最终的训练集,最后利用支持向量机对所选择的最具代表样本点进行训练建模。实验表明,所提方法可以大幅度降低支持向量机的学习代价,其分类精度比随机欠采样更优,而且可以达到采用完整数据集训练所得的结果

关键词: 海量数据, 分类, 聚类, 重采样, 支持向量机