Maximum Difference Feature Selection Algorithm Based on MRMR Algorithm

When encountering data with high dimensionality and high redundancy between features, we can select the most representative features by feature selection. In this paper, a maximum difference feature selection algorithm based on MRMR algorithm is proposed to select features for high dimensional data. On the basis of MRMR algorithm, we introduce Wasserstein distance to calculate the degree of difference between the features, so that the selected feature subset has the characteristics of maximum relevance, minimum redundancy, and maximum difference, and comparison experiments are carried out with classical feature selection algorithms such as MRMR. The experimental results show that in the vast majority of cases, the maximum difference feature selection algorithm in this paper is able to reduce the redundancy between features and improve the classification accuracy as the number of selected features increases.