Nature-inspired metaheuristics model for gene selection and classification of biomedical microarray data

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Gene expression datasets are vital for sample classification.
• Selecting a minimal set of informative genes is crucial for maximizing classification accuracy.
• Existing feature selection methods may suffer from premature convergence.

Purpose of the Study:

• To propose a hybrid machine learning framework for efficient gene selection.
• To optimize the balance between exploration and exploitation in search algorithms.
• To improve classification accuracy in gene expression data analysis.

Main Methods:

• A hybrid framework incorporating Cuckoo Search (CS) and Artificial Bee Colony (ABC) algorithms with Genetic Algorithms (GA).
• Independent Component Analysis (ICA) for initial gene extraction and feature reduction.
• Naive Bayes (NB) classifier with Leave-One-Out Cross-Validation (LOOCV) for performance evaluation.

Main Results:

• The proposed framework demonstrated a deeper search capability, avoiding premature convergence.
• Experimental results on six benchmark datasets showed superior performance compared to existing methods.
• The hybrid ICA and CS-based algorithm achieved higher classification accuracy with a smaller gene subset.

Conclusions:

• The proposed hybrid framework effectively identifies informative genes for enhanced sample classification.
• The integration of CS, ABC, and GA offers a robust approach to feature selection.
• This method provides a promising alternative for gene expression data analysis in bioinformatics.