一种支持向量机预处理方法的研究

Research on a Preprocessing Method of Support Vector Machine

摘要: 支持向量机(Support Vector Machine, SVM)在处理大规模数据集时，随着样本维度增高，样本数量增多会出现训练时间显著增多的问题。为了解决该问题，文章提出了一种基于主成分分析(Principal Component Analysis, PCA)和K边界近邻法(K Nearest Bound Neighbor, KNBN)的SVM预处理方法；先用PCA对训练数据降维消除训练数据中的冗余信息，然后利用KNBN预选取训练数据中的支持向量来减少训练数据量。数值实验结果表明，与PCA-SVM、KNBN-SVM和无数据预处理的SVM方法相比，采用本文提出的SVM预处理方法既保持了良好的分类预测精度，又缩短了大量训练时间。

Abstract: When the large-scale data set with higher dimension and larger number of samples is processed by the Support Vector Machine (SVM), the training time will increase significantly. In order to solve this problem, based on Principal Component Analysis (PCA) and K Nearest Bound Neighbor (KNBN), a preprocessing method of SVM is proposed. Firstly, PCA is used to reduce the dimension of the training data to eliminate the redundant information in the training data, and then KNBN is used to preselect the support vectors in the training data to reduce the amount of training data. The numerical experiment results show that SVM preprocessing method proposed in this paper, compared with PCA-SVM, KNBN-SVM and SVM without data preprocessing, can not only keep good classification prediction accuracy, but also save a lot of training time.