Leveraging Shapley Additive Explanations for Feature Selection in Ensemble Models for Diabetes Prediction

Area of Science:

• Medical Informatics
• Computational Biology
• Public Health

Background:

• Diabetes poses a global health challenge, exacerbated in India by lifestyle changes linked to urbanization.
• Effective diabetes management requires advanced prevention strategies and technological integration.

Purpose of the Study:

• To enhance the accuracy and efficiency of diabetes prediction models.
• To identify and validate the most influential features for diabetes risk using SHAP values.
• To assess the impact of feature selection on model performance.

Main Methods:

• Integration of Shapley Additive explanations (SHAPs) with ensemble machine learning models.
• Evaluation of model performance using three feature sets: all features, top three influential features, and features excluding the top three.
• Comparative analysis of ten different machine learning models.

Main Results:

• Models utilizing the top three most influential features demonstrated superior performance.
• The ensemble model achieved better performance across most metrics when focusing on the top features.
• Excluding the top three features resulted in a significant decrease in predictive performance.

Conclusions:

• Targeted feature selection using SHAP values is effective for improving diabetes prediction models.
• This approach enhances model efficiency and robustness for clinical applications.
• Identifying key predictive features is crucial for accurate and resource-efficient diabetes management.