一种基于因果网络的支持向量回归特征选择算法

A Support Vector Regression Feature Selection Method Based on Causal Network

为了提高支持向量回归算法的学习能力,提出了一种基于因果网络的特征选择算法.该方法假设目标变量和特征候选集之间符合一个因果网络模型,然后利用基于条件独立性测试的方法对目标变量的直接影响特征进行识别,从候选特征集之中获取与目标变量有着直接因果关系的特征子集.虚拟和真实数据集上的实验结果表明,该特征选择算法适用于支持向量回归算法,优于目前其他算法.

In order to improve Support Vector Regression （SVR） learning ability, a novel feature selection method based on causal network is proposed. Firstly, the target variable and its candidate feature set are assumed to conform a causal network model. Subsequently, the causal feature can be detected by conditional independence test based method. Both virtual and real experimental results show that the proposed algorithm outperforms other methods when applied to SVR.