A robust variable screening procedure for ultra-high dimensional data.

Area of Science:

• Statistics
• Computational Biology
• Genomics

Background:

• Variable selection is crucial in ultra-high dimensional regression.
• Existing methods like Sure Independence Screening (SIS) are sensitive to outliers, potentially causing inaccurate results, especially in small datasets.
• Robust pre-screening methods are needed to overcome these limitations.

Purpose of the Study:

• To develop a novel robust variable screening procedure for ultra-high dimensional data.
• To enhance the reliability of variable selection in the presence of outliers.
• To introduce Density Power Divergence-Sure Independence Screening (DPD-SIS) and its iterative extension.

Main Methods:

• Development of a new robust screening procedure based on Density Power Divergence (DPD) estimation.
• Introduction of DPD-SIS and iterative DPD-SIS.
• Extensive simulation studies to evaluate method performance.

Main Results:

• DPD-SIS and iterative DPD-SIS show superior performance compared to original SIS and other robust methods when outliers are present.
• The proposed methods are particularly effective in small samples with data contamination.
• Demonstrated utility in a real-world application concerning lipid metabolism regulation.

Conclusions:

• DPD-SIS offers a robust and reliable alternative for variable selection in ultra-high dimensional regression, especially when data contains outliers.
• The proposed methods mitigate the drawbacks of traditional SIS, leading to more accurate inference.
• This approach has significant implications for statistical modeling in fields prone to noisy data.