Ant colony optimization algorithm for interpretable Bayesian classifiers combination: application to medical...

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Biology

Background:

• Prediction models face challenges with performance in diverse environments and lack of interpretability.
• Ensemble classifiers improve performance but often reduce model interpretability.
• Interpretability is vital for users in fields like medicine and economics.

Purpose of the Study:

• To propose a novel classifier combination approach that enhances prediction performance and preserves model interpretability.
• To address the limitations of existing ensemble methods regarding explainability.
• To develop a solution tailored for Bayesian classifiers using Ant Colony Optimization.

Main Methods:

• Implementation of a novel classifier combination approach.
• Utilizing Ant Colony Optimization (ACO) for classifier selection and combination.
• Tailoring the ACO approach specifically for Bayesian classifiers.

Main Results:

• The proposed method successfully enhances overall prediction performance.
• Interpretability of the resulting classification model is preserved.
• Validation through case studies in the medical domain (heart disease, cardiotocography).

Conclusions:

• The novel approach effectively balances performance and interpretability in classification models.
• The Ant Colony Optimization-based method offers a viable solution for improving Bayesian classifiers.
• The technique is particularly valuable in critical decision-making domains like healthcare.