基于概率密度距离的监督特征选择

Supervised feature ranking approach based on probability density interval

为了降低特征维数,提高分类效率,提出了一种新的基于概率密度距离的有监督的特征排序方法。首先依次对所有样本的某一维特征进行加权变换,然后对变换后的各类别样本进行概率密度估计,计算由该特征加权变换后所引起的各类别样本的类间概率密度距离,距离越大,则该特征对于区分各类别样本的作用越大,以此来对特征进行排序。实验结果表明,该方法是有效的,而且表现出了比经典的Relief-F特征排序方法更好的性能。

In order to reduce dimensionality and improve efficiency, a novel supervised feature ranking approach based on probability density interval is proposed. First one of the dataset＇ s feature is weighted, then calculate the probability density interval between classes. The most important feature will result the biggest distances between classes. Therefore, probability density interval of inter-class to rank feature is used. Several experimental results demonstrate the effectiveness and the advantage of our approach here over Relief-F.