Refactoring and Benchmarking Relief-Based Feature Selection Algorithms in scikit-rebate for Genomic Data

Forum topic · 小凯 · 2026-09-01

Summary

A new study (arXiv:2608.28552) by Kazemi-Nia, Bandhey, Freda, and Urbanowicz refactors, optimizes, and expands the scikit-rebate Python package of Relief-based algorithms (RBAs), which are filter-based feature selection methods capable of detecting feature interactions—something most filters cannot do—while remaining far cheaper than wrapper or embedded methods. The authors add SWRF*, mu-Relief, and five novel RBA variants implementing alternative neighbor-selection and feature-scoring strategies, then benchmark all algorithms on simulated genomic datasets varying sample size, feature count, heritability, and underlying association type (main effects, 2-way, and 3-way interactions). Key findings: most RBAs detect 2-way interactions in noisy data; RBAs using 'far' scoring (best: MultiSWRFDB*) excel at 2-way interaction detection but are less sensitive to main effects; SWRF, MultiSWRF, MultiSURF, and MultiSWRFDB perform best on main-effect and 2-way data, with MultiSWRFDB best when 3-way interactions are considered. The refactored codebase runs 10 to 35 times faster. The new RBAs robustly preserve main effects and epistatic 2-way interactions, retaining predictive signal for downstream modeling of high-dimensional biomedical data.

论文概要 (Paper Overview)

Field: Machine Learning Authors: Kia Kazemi-Nia, Harsh Bandhey, Philip J. Freda, Ryan J. Urbanowicz Published: 2026-08-28 arXiv: 2608.28552

Abstract

As a precursor to high-dimensional biomedical data modeling, reliable feature selection can reduce computational expense, improve modeling performance, and yield simpler, more interpretable models. However, most filter-based feature selection methods struggle to detect feature interactions, while wrapper or embedded feature selection methods are computationally expensive. Relief-based algorithms (RBAs) are filter methods that are sensitive to feature interactions while mitigating these other limitations.

This study (1) refactors, optimizes, and expands the scikit-rebate Python package with existing and newly proposed RBA variants, and (2) conducts rigorous RBA benchmark comparisons across diverse genomic simulations. The expanded scikit-rebate includes SWRF*, mu-Relief, and 5 novel RBA variants implementing alternative strategies for neighbor selection and feature scoring.

Key Findings

All RBAs were evaluated on simulated genomic datasets varying in sample size, feature count, heritability, and underlying association type (main effects and interactions), comparing predictive feature rankings and runtime.
Except for mu-Relief, all RBAs excelled at detecting 2-way interactions in noisy data.
RBAs using 'far' scoring were best at detecting 2-way interactions — MultiSWRFDB* performed best — but were considerably less sensitive to main effects.
SWRF, MultiSWRF, MultiSURF, and MultiSWRFDB performed best on datasets with main effects and 2-way interactions; MultiSWRFDB was best when 3-way interactions were also considered.
The refactoring of scikit-rebate reduced RBA runtimes by 10 to 35 times.
The newly introduced RBAs are among the strongest performers, robustly preserving main effects and 2-way epistatic (interacting) signals for downstream modeling.

Conclusion

The refactored and expanded scikit-rebate package offers faster, interaction-sensitive filter-based feature selection. The new RBA variants balance detection of main effects and 2-way epistatic interactions, preserving predictive signal for downstream modeling of high-dimensional biomedical (genomic) data.

--- *Auto-collected on 2026-09-01.*