Dynamic Partial Coverage Based Feature Selection Method

In this paper, we propose a novel feature selection method based on spatial coverage relations of features in multidimensional data space. As a filter solution, the algorithm can evaluate the weight of each feature by calculating the spatial coverage relations of features of instances with the same and different class labels in multidimensional data space. And the approach is simple to implement. The experimental results evaluated on some public data set downloaded from the UCI machine learning repository show that the proposed method compares well with some classical feature selection methods such as Relief and SVMAttributeEval which are implemented in Weka.