基于负相关学习的支持向量机集成算法

提出一种基于负相关学习理论的支持向量机集成方法，该方法能有效解除各支持向量机之问的相关性，使得集成个体有较大的差异。并采用了演化策略对支持向量机的核函数和相关参数白适应地进行选择。仿真实验表明．该方法不仅能有效地解决支持向量机模型选择的难题，而且能以很小的代价显著提高学习系统的泛化能力。

Algorithm of Support Vector Machine Ensemble Based on Negative Correlation Learning

This paper presents a constructive algorithm for training negative correlation support vector machines (SVMs) ensembles. This approach can produce individual SVMs whose errors tend to be negatively correlated, so the diversity is emphasized among individual SVMs in an ensemble. Kernel function and its parameters of individual SVMs can be selected automatically with evolutionary strategy. Simulation shows that this method can not only solve model selection problem of SVM, but also improve SVM generalization performance effectively with small cost.