Ensemble Feature Selection for Microarray Data Classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Microarray data classification faces challenges due to high dimensionality and small sample sizes, leading to unstable feature selection.
• Existing ensemble feature selection methods often fail to effectively balance feature diversity and quality.

Purpose of the Study:

• To introduce a novel Ensemble Feature Selection Method (EFSM) designed to enhance feature selection stability and performance in high-dimensional microarray data.
• To address the limitations of traditional methods in balancing feature diversity and predictive accuracy.

Main Methods:

• EFSM utilizes randomized neural networks to create diverse non-linear feature mappings (views) for generating a robust candidate pool of feature selectors.
• A novel ensemble pruning technique is employed, formulated as a Semi-Definite Programming (SDP) problem to optimize individual selector accuracy and pairwise diversity.
• Feature rankings are aggregated using the Borda count method.

Main Results:

• EFSM demonstrated superior and stable performance across 15 biological datasets compared to nine state-of-the-art feature selection methods.
• The method achieved improved classification accuracy when tested with popular classifiers on high-dimensional data.

Conclusions:

• The proposed EFSM offers a robust and effective solution for feature selection in high-dimensional microarray data analysis.
• EFSM successfully balances feature diversity and quality, leading to enhanced classification performance and stability.