Integrating Machine Learning with Explainable AI in Healthcare Analytics for Diabetes Prediction

This study introduces an innovative and comprehensive framework for predicting diabetes, skillfully integrating classical machine learning models with explainable artificial intelligence (XAI) techniques. This integration not only enhances predictive accuracy but also significantly improves clinical interpretability, making the results more accessible to healthcare professionals. To achieve this, we utilized a carefully constructed synthetic electronic medical records (EMR) dataset that closely replicates real-world trends associated with Type I diabetes. Our analysis involved training logistic regression and random forest models, demonstrating competitive performance levels. Distinct from previous research that often prioritizes accuracy at the expense of interpretability, our approach uniquely harnesses the capabilities of SHAP and ELI5 to deliver detailed, feature-specific explanations for each model prediction. This dual focus not only provides insights into the critical risk factors influencing diabetes-such as glucose levels, body mass index (BMI), and age-but also tackles the potential biases that can arise from using synthetic data. The findings from our study present a replicable framework that effectively bridges the divide between highperformance predictive modeling and the pressing need for clear, transparent insights in healthcare analytics. The implications of our research aim to empower practitioners with actionable information that can enhance patient care and outcomes in diabetes management.