构造特征复杂性减低的支持向量机

支持向量机(SVM)较一般的机器学习方法显示出更好的泛化能力.然而,在实际的数据中经常存在着大量冗余、噪声或者不可靠的特征,这严重影响到SVM的性能.因此,有必要减低特征复杂性以获取更好的SVM结果.本文提出了一种基于遗传算法(GA)的嵌入式框架下的特征优化算法,以构造改进SVM.针对选择的UCI成人数据库的实验表明,与原始的SVM相比,提出的改进SVM方法获得了更少的支持向量数目和更好的分类精度.

Building Support Vector Machine with Reduced Feature Complexity

Support Vector Machine (SVM) has revealed better generalization than conventional machine learning methods. However, in the real data there often exist a large number of redundant, noisy or unreliable features to deteriorate the function of SVM strongly. So to reduce the feature complexity, it is necessary to improve the performance of SVM for better results. A method to build modified SVM, which is based on embedded methods for feature optimization using Genetic Algorithm (GA),is proposed in this paper. The experimental results on selected UCI Adult data base show that compared with original SVM classifier, the number of support vector decreases and better classification results are achieved based on our modified SVM.