A voting-based machine learning approach for classifying biological and clinical datasets.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning Applications

Background:

• Existing machine learning methods for biological/clinical data classification face limitations like overfitting and poor performance on large datasets.
• Feature selection is often overlooked in preprocessing, hindering model generalization.
• There is a need for robust machine learning frameworks to handle complex biological data.

Purpose of the Study:

• To introduce a novel machine learning framework for improved biological/clinical data classification.
• To address limitations of existing methods, including overfitting and performance degradation on large datasets.
• To enhance feature selection and classification accuracy in biological data analysis.

Main Methods:

• An optimization algorithm (Trader) was extended for near-optimal feature/gene subset selection.
• A voting-based framework was developed for high-accuracy classification of biological/clinical data.
• The proposed framework was evaluated on 13 diverse biological/clinical datasets.

Main Results:

• The Trader algorithm identified near-optimal feature subsets with statistical significance (p < 0.01).
• The machine learning framework achieved approximately 10% improvement in accuracy, precision, recall, specificity, and F-measure on large datasets.
• Performance gains were validated through fivefold cross-validation, demonstrating robustness.

Conclusions:

• Effective configuration of algorithms enhances machine learning prediction power in biological sciences.
• The proposed framework offers a practical approach for developing diagnostic healthcare systems.
• This research aids in designing effective treatment plans through improved data analysis.