一种新的软间隔支持向量机分类算法

New support vector machine based classification algorithm and its application

软间隔支持向量机(SVM)分类算法是目前最具有代表性的模式分类算法之一,它在应用中的一个主要困难是确定控制参数C。提出一种新的软间隔SVM分类算法,通过松弛变量改变约束条件,允许数据点进入分离区域但不越过分类超平面,从而避免了参数C的确定问题。计算机实验和故障诊断实例表明,基于新算法的SVM分类器有较高的分类准确性和较好的泛化性能,能够实际应用于模式分类。

Vapnik＇s support vector machine （SVM） based soft-margin classifier is one of the most popular pattern classification models in use. The main difficult work in this SVM classifier is to determine the regulation parameter C. A proposes a new soft-margin SVM classification algorithm was proposed, which could avoid the design problem of parameter C. The main idea of the new classification algorithm is to permit data-points located in the separated area but not cross the hyper-plane through modifying the restriction conditions with slack variables. The simulation examples and fault diagnosis results show that the new algorithm can make SVM classifier have good generalization performance.