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,以构造改进的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． Support Vector Machine (SVM) shows better generalization ability than general machine learning methods. However, there is often a large amount of redundant, noisy or unreliable features in the actual data, which seriously affects the performance of the SVM. Therefore, it is necessary to reduce the complexity of the features for better SVM results. This paper presents a feature-based optimization algorithm based on genetic algorithm (GA) embedded framework to construct an improved SVM. Experiments on selected UCI adult databases show that the proposed improved SVM approach achieves fewer support vectors and better classification accuracy than the original SVM.