A Bootstrap Framework for Aggregating within and between Feature Selection Methods.

Area of Science:

• Data Science
• Machine Learning
• Bioinformatics

Background:

• Big data applications are increasing, leading to noisy and redundant datasets.
• Feature selection is crucial for managing complex datasets but different methods yield varied results.
• Aggregating feature selection results enhances subset diversity and reliability.

Purpose of the Study:

• To introduce a general framework for ensembling multiple feature selection methods.
• To address concerns regarding the variability of results from different feature selection approaches.
• To improve the understanding and stability of feature selection processes in big data.

Main Methods:

• Developed a framework for ensemble feature selection.
• Generated diversified datasets from original observations.
• Aggregated feature importance scores using Within Aggregation Method (WAM) and Between Aggregation Method (BAM).

Main Results:

• WAM proved effective in identifying the best feature selection method for a given dataset.
• WAM demonstrated superior stability compared to BAM in identifying important features.
• Both WAM and BAM showed comparable computational demands across 13 real datasets.

Conclusions:

• The proposed framework offers a robust approach to feature selection in big data.
• WAM is a valuable tool for selecting optimal feature selection algorithms.
• Employing both WAM and BAM enhances practitioners' understanding of the feature selection landscape.