Feature Selection Based on Grouped Sorting
Zhigang Shang, Mengmeng Li · 2016
As an effective dimensionality reduction method, feature selection can remove the irrelevant variables and increase the accuracy in machine learning. In this paper, a feature selection method based on grouped sorting is proposed to solve the problem of high-dimensional data processing. As in this work, feature grouping is firstly carried out with the redundancy between the features as the group criteria and then we sort the features in each group by the classifying capacity. Finally we select the feature of the front rank in each group to constitute a new feature space. Through this, the redundant and irrelevant features are removed. We test the proposed method in numerical experiments on several data sets by different typical classifiers. The experimental results suggest that the method is effective.