Robust feature selection for cancer microarray data using a hybrid mRMR and Binary Lion Optimization Algorithm

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Oncology

Background:

• Cancer microarray datasets often contain numerous irrelevant or noisy features, hindering classification accuracy.
• Feature selection is crucial for improving microarray analysis by identifying valuable features.
• Existing optimization methods can struggle with the NP-hard nature of feature selection, leading to local optima.

Purpose of the Study:

• To develop and evaluate a novel Binary Lion Optimization (BLO) algorithm for effective feature selection in cancer microarray datasets.
• To address the limitations of continuous optimization in existing Lion Optimization (LO) methods for discrete feature selection tasks.
• To enhance classification performance by identifying optimal feature subsets in high-dimensional cancer data.

Main Methods:

• A wrapper-based Binary Lion Optimization (BLO) algorithm was developed using an S-shaped Transfer Function.
• Minimum Redundancy Maximum Relevance (mRMR) was employed as a pre-processing filter for dimensionality reduction.
• The mRMR-BLO approach was tested on 11 benchmark cancer microarray datasets and compared against four other binary optimization techniques.

Main Results:

• The proposed mRMR-BLO algorithm achieved the highest prediction accuracy compared to existing methods.
• Effective feature selection was demonstrated, resulting in smaller, more informative feature sets.
• The algorithm showed strong performance across various cancer types and high-dimensional data.

Conclusions:

• The mRMR-BLO algorithm offers a robust and efficient solution for feature selection in cancer microarray analysis.
• This approach enhances classification accuracy and reduces computational complexity by identifying optimal feature subsets.
• BLO presents a promising metaheuristic for tackling NP-hard optimization problems in bioinformatics.