Abstract: Model combination is an important approach to improving the generalization performance of Support Vector Machine（SVM）, but usually has low computational efficiency. An effective SVM model combination method based on Bayesian model averaging over regularization path is presented. The initial candidate model set is constructed using regularization path algorithm, and the probabilistic interpretation of Support Vector Machine is described. The SVM prior is simply a Gaussian Process（GP） over the models, while model posterior probability is obtained by Bayesian formula, and the combination on the initial models is achieved using Bayesian model averaging. Comparative experiments on benchmark databases against common cross-validation model selection methods and Wahba’s Generalized Approximate Cross-Validation（GACV） show that the proposed combination algorithm is effective and efficient.

Key words: Support Vector Machine（SVM）, model combination, regularization path, Gaussian process, Generalized Approximate Cross-Validation（GACV）

摘要： 模型组合是提高支持向量机泛化性的重要方法，但存在计算效率较低的问题。提出一种基于正则化路径上贝叶斯模型平均的支持向量机模型组合方法，在提高支持向量机泛化性的同时，具有较高的计算效率。基于正则化路径算法建立初始模型集，引入对支持向量机的概率解释。模型的先验可看做是一个高斯过程，模型的后验概率通过贝叶斯公式求得，使用贝叶斯模型平均对模型进行组合。在标准数据集上，实验比较了所提出的模型组合方法与交叉验证及广义近似交叉验证（GACV）方法的性能，验证所提出的模型组合方法的有效性。

关键词: 支持向量机, 模型组合, 正则化路径, 高斯过程, 广义近似交叉验证（GACV）