Unsupervised feature selection with robust data reconstruction (UFS-RDR) and outlier detection

作者：

Highlights：

• Feature selection methods in unsupervised learning are sensitive to outliers.

• A novel unsupervised feature selection with robust data reconstruction is proposed.

• It minimizes the graph regularized weighted data reconstruction error function.

• This function down-weights the clustering observations having large distances.

• The proposed method outperform the competitive methods in the presence of outliers.

摘要

•Feature selection methods in unsupervised learning are sensitive to outliers.•A novel unsupervised feature selection with robust data reconstruction is proposed.•It minimizes the graph regularized weighted data reconstruction error function.•This function down-weights the clustering observations having large distances.•The proposed method outperform the competitive methods in the presence of outliers.

论文关键词：Unsupervised feature selection,Data reconstruction,Outliers,Mahalanobis distance,Robustness

论文评审过程：Received 12 January 2021, Revised 1 February 2022, Accepted 26 March 2022, Available online 6 April 2022, Version of Record 19 April 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117008