Explainable Machine Learning Models for Type-2 Diabetes Prediction

Type-2 Diabetes Mellitus (T2DM) poses a major challenge for the UK's NHS. Traditional diagnostic systems lack scalable integration of behavioural and demographic predictors. This study develops explainable tree-based ensemble models (RF, XGBoost, LightGBM, CatBoost, and a stacked ensemble) using the BRFSS-2015 dataset to enhance transparency and predictive reliability. Our methodology integrates rigorous preprocessing, evaluation using accuracy, and application of k -fold cross validation to assess model robustness. Preliminary findings suggest that these models have achieved competitive results by application of cross validation by reaching accuracy of 93.13 %, as compared with existing literature, while maintaining interpretability through feature importance and LIME analysis. The study highlights the balance between predictive performance and explainability, emphasising the potential of ML in augmenting NHS practices for early detection and personalised care of diabetes patients.