Variable Selection in the Presence of Missing Data: Imputation-based Methods.

Area of Science:

• Statistics
• Data Science
• Machine Learning

Background:

• Variable selection is vital for regression analysis and predictive accuracy.
• Existing methods are well-developed for complete data but require adaptation for missing data.
• Handling missing data is essential for reliable statistical modeling.

Purpose of the Study:

• To explore statistical methods for variable selection in the presence of missing data.
• To investigate the integration of imputation techniques with variable selection strategies.
• To highlight the need for robust methods under missing at random (MAR) and missing completely at random (MCAR) assumptions.

Main Methods:

• Review of three main strategies for variable selection with imputation.
• Strategy 1: Applying selection methods to imputed datasets and combining results.
• Strategy 2: Applying selection methods to stacked imputed datasets.
• Strategy 3: Combining imputation with resampling techniques like bootstrap.

Main Results:

• Imputation-based variable selection methods are of particular interest.
• Current methods are valid under MAR and MCAR assumptions.
• The field of variable selection with missing data is still developing.

Conclusions:

• Variable selection methods must account for missing data mechanisms.
• Imputation offers a practical approach for handling missing data in variable selection.
• Further research is needed to advance variable selection techniques for incomplete datasets.