OmniGA: Optimized Omnivariate Decision Trees for Generalizable Classification Models.

Area of Science:

• Machine Learning
• Computational Biology
• Data Science

Background:

• Classification tasks present challenges due to varying complexity, data size, and class imbalance.
• Selecting optimal models and parameters for high performance in classification is a non-trivial problem.
• There is a continuous need for robust and efficient models applicable to diverse datasets.

Purpose of the Study:

• To introduce OmniGA, a novel framework for optimizing omnivariate decision trees.
• To leverage parallel genetic algorithms, deep learning structures, and ensemble methods for enhanced classification.
• To demonstrate the superior performance of OmniGA on various datasets, particularly those from biomedical domains.

Main Methods:

• OmniGA framework combines parallel genetic algorithms with deep learning and ensemble learning.
• Omnivariate decision trees are optimized using the proposed OmniGA approach.
• Performance evaluation was conducted on 12 diverse datasets, primarily from biomedical research.

Main Results:

• OmniGA systematically outperformed commonly used machine learning models on all tested datasets.
• The framework achieved a reduction in F1 score error ranging from 2.25% to 100% compared to the best baseline model.
• Results indicate that OmniGA generates robust classification models with significantly improved performance.

Conclusions:

• OmniGA offers a robust and efficient solution for complex classification problems.
• The framework demonstrates superior performance and improved accuracy in diverse applications, especially in biomedical data analysis.
• The developed OmniGA code and datasets are publicly available for further research and application.