Diabetes, a global health concern, demands enhanced prediction accuracy for timely intervention. Traditional methods, reliant on patient history and diagnostic tests, can be subjective and prone to error. This study leverages machine learning (ML) to improve diabetes prediction using data from the Pima Indians Diabetes Database (PIDD) and addresses the “black-box” challenge posed by complex ML models through Explainable Artificial Intelligence (XAI). We utilize Local Interpretable Model-agnostic Explanations (LIME) to enhance model transparency and compare its local explanations with global feature rankings derived via Recursive Feature Elimination (RFE). Our findings show that RFE consistently identifies features such as age, BMI, number of pregnancies, insulin levels, and glucose concentration as critical, while LIME emphasizes glucose concentration, BMI, insulin levels, blood pressure, and diabetes pedigree function in individual predictions. This integrated approach enhances both model interpretability and clinical trust, advancing the application of AI in diabetes management.