Network-based dimensionality reduction of high-dimensional, low-sample-size datasets
作者：
Highlights：
• A novel network-based nonparametric method (NDA) is proposed to perform dimensionality reduction.
• NDA finds latent variables (LVs) by community detection of the correlation graph of indicators.
• NDA provides feature selection, ignoring indicators with low and common communalities.
• NDA provides both the set of LVs and the set of indicators belonging to the LVs.
• NDA is tested and compared with both principal component analysis and factoring analysis on publicly available databases.
摘要
•A novel network-based nonparametric method (NDA) is proposed to perform dimensionality reduction.•NDA finds latent variables (LVs) by community detection of the correlation graph of indicators.•NDA provides feature selection, ignoring indicators with low and common communalities.•NDA provides both the set of LVs and the set of indicators belonging to the LVs.•NDA is tested and compared with both principal component analysis and factoring analysis on publicly available databases.
论文关键词：Nonparametric methods,Dimensionality reduction,Community detection,Communality analysis
论文评审过程：Received 9 June 2021, Revised 3 March 2022, Accepted 30 May 2022, Available online 4 June 2022, Version of Record 17 June 2022.
论文官网地址：https://doi.org/10.1016/j.knosys.2022.109180
