TS-RePSO: A Three-Stage Feature Selection Method Combing ReliefF and PSO in Bioinformatics.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• High-dimensional biomedical data presents challenges due to feature redundancy, known as the curse of dimensionality.
• Existing two-stage (filter-wrapper) and one-stage feature selection methods struggle with threshold setting and can get stuck in local optima.

Purpose of the Study:

• To propose a novel three-stage feature selection method, TS-RePSO, to overcome limitations of existing approaches.
• To enhance feature selection accuracy and efficiency in high-dimensional biomedical datasets.

Main Methods:

• The proposed TS-RePSO method integrates ReliefF for feature weighting and sorting (filter stage).
• A density equalization strategy is used for grouping ranked features (grouping stage).
• A modified Particle Swarm Optimization (PSO) algorithm searches grouped features using in-group and out-group evaluation (wrapper stage).

Main Results:

• Extensive experiments on 5 benchmark and 6 real-world datasets were conducted.
• The TS-RePSO method demonstrated superior performance compared to existing feature selection techniques.
• The proposed grouping PSO effectively searched for optimal feature subsets.

Conclusions:

• The three-stage TS-RePSO method effectively addresses the curse of dimensionality in biomedical data.
• TS-RePSO offers an improved approach to feature selection, enhancing performance and overcoming local optima issues.