Biochemical biomarker-Driven deep learning framework with SHAP-based feature interpretation for diabetes...

Area of Science:

• Endocrinology and Metabolism
• Computational Biology
• Public Health

Background:

• Diabetes mellitus is a growing global health concern, characterized by impaired insulin production or utilization.
• Progression from normal glucose regulation to prediabetes and type 2 diabetes necessitates early detection to prevent severe complications.
• Increasing diabetes prevalence is linked to lifestyle factors, environmental pressures, and socioeconomic determinants.

Purpose of the Study:

• To develop and validate a predictive model for the early identification of diabetes.
• To enhance model interpretability by identifying key predictive features.
• To improve upon existing diagnostic methods for diabetes.

Main Methods:

• Integration of a Deep Neural Network with feature ranking and a statistical algorithm.
• Application of Shapley Additive exPlanations (SHAP) for feature interpretability.
• Rigorous evaluation using 10-fold cross-validation.

Main Results:

• The proposed model achieved an average accuracy of 95.72%.
• Demonstrated significant improvements compared to traditional machine learning models and recent benchmarks.
• Identified key influential features for diabetes prediction through SHAP analysis.

Conclusions:

• Advanced analytical tools, like the proposed predictive model, are crucial for effective early diabetes screening.
• Socioeconomic factors such as urbanization, dietary shifts, and healthcare access play a significant role in diabetes prevention and management.
• Early detection and intervention are vital to mitigate long-term diabetes complications.