An Efficient hybrid filter-wrapper metaheuristic-based gene selection method for high dimensional datasets.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• High-dimensional datasets in gene expression analysis often lead to overfitting in classification models.
• Existing feature selection methods struggle with instability, long convergence times, and suboptimal solutions.
• Effective feature extraction is crucial for accurate and cost-efficient data classification.

Purpose of the Study:

• To propose a hybrid feature selection method for identifying effective features in large-scale gene datasets.
• To enhance classification accuracy and reduce the cost of data classification.
• To address the limitations of existing feature selection techniques, including instability and semi-optimal solutions.

Main Methods:

• A two-phase hybrid approach combining filtering and wrapping techniques.
• Phase 1 (Filtering): Feature weighting using the Relief method.
• Phase 2 (Wrapping): Feature subset search using the Shuffled Frog Leaping Algorithm (SFLA) and the IWSSr algorithm.

Main Results:

• The proposed hybrid method achieved a more compact set of features compared to existing methods.
• Experimental results on standard gene expression datasets demonstrate high classification accuracy.
• The approach effectively extracts relevant features, mitigating overfitting in high-dimensional data.

Conclusions:

• The hybrid IWSSr and SFLA-based feature selection method offers superior performance for gene expression data.
• The proposed technique provides a robust solution for selecting optimal feature subsets, improving classification efficiency and accuracy.
• The study's findings contribute to more effective analysis of complex biological datasets.