Grouped feature screening for ultra-high dimensional data for the classification model

Abstract

Existing model-free ultra-high-dimensional feature screening methods mainly focus on the individual covariate. However, many variables have a community structure, such as grouped covariates in which all variates have high correlation and some associations in one group. There are some research studies about grouped feature screening for ultra-high-dimensional data for a linear model by promoting sure independence screening and high-dimensional ordinary least-squares projector to the group version. Hence, we propose a new ultra-high-dimensional grouped feature screening method for a classification model, which is model-free and suitable for continuous and categorical covariates. Compared with individual covariate feature screening, the proposed method yields a better screening performance and classification accuracy. The grouped feature screening and ranking consistency properties of the proposed method are established. We illustrate the finite sample performance of the proposed method by simulation and real data analysis.

KEYWORDS:

• Grouped feature screening
• sure screening property
• information gain ratio
• joint entropy
• multi-class classification

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The lung biological data used in the real data analysis were obtained from the feature selection database of Arizona State University (http://featureselection.asu.edu/).

Additional information

Funding

This work was supported by National Natural Science Foundation of China [grant number 71963008] and Natural Science Foundation of Guangxi China [grant number 2018GXNSFAA294131].