基于多分类支持向量机的模式识别研究

Pattern Recognition Based on Multi-class Support Vector Machine

支持向量机是现代人工智能领域中的一个重要分支,它在统计学习理论的基础上,实现了结构风险最小化,提高了分类器的泛化能力,保证了分类的准确度。论文提出一种基于多分类支持向量机的模式识别方法,采用特征选择序列极小化算法对数据样本特征进行选择,并在此基础上,分析对比了"一对一"分类算法和"一对多"分类算法,实验结果表明,"一对一"分类算法的分类准确性较高,且具有较好的推广能力。

Support vector machine is an important branch in the field of modem artificial intelligence. Based on statistical learning theory, it implements the structural risk minimization, improves the generalization ability of the classifier, guarantees the accuracy of classification. This paper proposes a pattern recognition method based on support vector machine The eigenval- ues selection sequence minimizing method is adopted to select the characteristic of data sample. On this basis, the ＂one-against- one＂ classification algorithm and ＂one-against-all＂ classification algorithm are analyzed and compared. The experimental results show that the ＂one-against-one＂ classification algorithm has higher accuracy, and better generalization ability.