Interpretable Machine Learning for Diabetes Prediction

Background

Diabetes is a growing global health challenge, affecting over 14% of U.S. adults as of 2022. Our project applies machine learning to health data to better understand and predict diabetes risk. Unlike traditional approaches that optimize performance on a single dataset, we focus on comparing consistent models across different datasets to uncover how data quality and composition affect model outcomes and interpretability.

Goals

• Use interpretable machine learning models to predict diabetes diagnoses

• Compare model performance across clinical and survey-based datasets

• Highlight both global and dataset-specific risk factors

• Make results accessible to healthcare professionals via SHAP explanations