基于Bagging集成的高维不平衡数据特征选择方法OA北大核心CHSSCDCSSCICSTPCD

With the development of big data,samples in many application areas are presented in high-dimensional forms,and the high-dimensional characteristics of datasets will attenuate the classification effect of imbalanced learning.Aiming at the classification of high-dimensional imbalanced data,this paper proposes an adaptive feature selection method WAFS based on SVM-RFE and Bagging ensemble,which combines embedded and wrapper feature selection methods to adaptively select the opti-mal features to form feature space.Through 5 high-dimensional imbalanced public datasets with different dimensions(100～25000),WAFS is compared with the filter-based CSS feature selection algorithm and the embedded ASG feature selection algo-rithm.Also,the optimal sampling method for different datasets and the optimal rate of feature space of datasets with different di-mension are explored.AUC,Acc,Recall,F1-score and G-mean is taken as the evaluation indicators,and the experiment is con-ducted to show that the WAFS algorithm has good performance on datasets with different dimensions,especially in high-dimen-sional and imbalanced datasets with small samples,and that the model has strong stability and generalization under the premise of ensuring accuracy.