Feature Screening for High-Dimensional Variable Selection in Generalized Linear Models.

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• Existing two-stage feature screening methods primarily focus on linear models.
• High-dimensional data analysis requires efficient methods for dimension reduction and feature selection.
• Generalized linear models are widely used but less explored for feature screening.

Purpose of the Study:

• To extend the sure independence screening method to generalized linear models, particularly for binary responses.
• To develop a computationally efficient and accurate two-stage feature screening method.
• To introduce the point-biserial sure independence screening (PB-SIS) method.

Main Methods:

• A two-stage approach involving initial dimension reduction and subsequent penalized methods.
• Utilizing point-biserial correlation for feature screening in generalized linear models with binary outcomes.
• Developing the point-biserial sure independence screening (PB-SIS) algorithm.

Main Results:

• PB-SIS demonstrates high efficiency and accuracy in feature screening.
• The method exhibits the sure independence property under specific conditions.
• Simulation studies confirm the effectiveness, accuracy, and efficiency of PB-SIS.

Conclusions:

• PB-SIS is an effective feature screening method for high-dimensional generalized linear models.
• The method offers a valuable tool for analyzing complex datasets in various scientific fields.
• PB-SIS provides a robust solution for feature selection with improved computational performance.