Benchmark of filter methods for feature selection in high-dimensional gene expression survival data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional data analysis is essential in genomics.
• Feature selection is critical for survival data analysis.
• Benchmark studies for gene expression survival data are scarce.

Purpose of the Study:

• To compare 14 filter methods for feature selection on high-dimensional gene expression survival data.
• To provide guidance on selecting appropriate filter methods.
• To evaluate performance based on predictive accuracy, runtime, and stability.

Main Methods:

• Utilized 11 high-dimensional gene expression survival datasets.
• Implemented and assessed 14 distinct filter methods.
• Analyzed predictive model accuracy, computational runtime, feature count, and selection stability.

Main Results:

• The simple variance filter demonstrated superior performance.
• Correlation-adjusted regression scores filter emerged as a viable alternative.
• Identified groups of filter methods with similar ranking behaviors.

Conclusions:

• The variance filter is a highly effective and efficient method for feature selection in this context.
• Researchers can confidently select the variance filter for its performance and simplicity.
• Alternative methods like correlation-adjusted regression scores offer comparable accuracy with more complexity.