代价敏感相关向量机

Cost Sensitive Relevance Vector Machine

相关向量机(RVM)是在稀疏贝叶斯框架下提出的稀疏模型,由于其强大的稀疏性和泛化能力,近年来在机器学习领域得到了广泛研究和应用,但和传统的决策树、神经网络算法及支持向量机一样,RVM不具有代价敏感性,不能直接用于代价敏感学习。针对监督学习中错误分类带来的代价问题,提出代价敏感相关向量分类(CS-RVC)算法,在相关向量机的基础上,通过赋予每类样本不同的误分代价,使其更加注重误分类代价较高的样本分类准确率,使得整体误分类代价降低以实现代价敏感挖掘。实验结果表明,该算法具有良好的稀疏性并能够有效地解决代价敏感分类问题。

Relevance Vector Machine （ RVM） is a sparse model proposed on the basis of sparse Bayesian framework, it has been widely studied and applied in the field of machine learning in recent years because of its strong sparsity and generalization ability. However, like the traditional decision tree, neural network algorithm and support vector machine, RVM does not have the ex-pense of sensitivity, can not be directly used for cost-sensitive learning.To deal with the cost sensitive problem brought by mis-classification in supervised learning, cost-sensitive relevance vector classification（ CS-RVC） algorithm was proposed by integrating misclassification cost of each type sample based on RVM.Experiments show that CS-RVC has good sparsity and can effectively solve the problem of cost-sensitive classification.