Improving Type 2 Diabetes Prediction: Comparative Evaluation of Machine Learning Classifiers Using Balanced Data from...

Area of Science:

• Public Health
• Machine Learning
• Epidemiology

Background:

• Type 2 diabetes mellitus (T2DM) is a growing health issue in Africa, with limited region-specific predictive models.
• Existing research faces challenges like data leakage, class imbalance, and overfitting, hindering clinical use, especially in digital health.

Purpose of the Study:

• To develop and validate machine learning (ML) prediction models for T2DM tailored to African populations.
• To address methodological limitations in previous studies and enable deployment in digital health contexts.

Main Methods:

• Analysis of data from 2,010 participants in the H3Africa AWI-Gen cohort in Ghana.
• Application of rigorous preprocessing, including SMOTE for class imbalance and exclusion of leakage-prone biomarkers.
• Evaluation of eight ML classifiers with Bayesian hyperparameter optimization and 5-fold cross-validation.

Main Results:

• The optimized XGBoost model achieved an AUC of 0.845.
• Excluding glucose as a predictor was crucial to avoid biased evaluation.
• Models using anthropometric and lifestyle data (AUC = 0.783) showed strong predictive ability, with waist circumference, physical activity, and BMI being key predictors.

Conclusions:

• ML models using routine clinical and lifestyle data can achieve clinically relevant T2DM prediction for African digital health applications.
• This study provides a robust, data-driven framework for early T2DM detection, addressing prior methodological gaps.
• Findings have implications for public health policy and digital screening programs in similar settings.