Interpreting Blood Glucose Effects in Juvenile Diabetes Through XAI

This study reveals that leveraging XAI to assess the impact of nutritional and contextual parameters on blood sugar for type 1 diabetes mellitus in machine learning methods. The method of this study includes Type 1 diabetes mellitus (T1DM) is defined by insulin dependent diabetes as well as difficulty in controlling blood sugar. This project introduces a blood glucose forecasting model that utilizes machine learning methods, namely the Random Forest algorithm, to predict blood glucose levels from past patient data. The model utilizes different input features, such as past glucose levels, insulin doses, and carb intake, to make precise short-term predictions. Data preprocessing methods are used enhance the dataset quality. The model is gauged as effective through its performance measures in terms of Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), reflecting that the model could effectively predict high variability in blood glucose with utmost precision. Last but not least, the adoption of XAI such as SHAP and LIME adds interpretability to this model and helps users to trace in what manner the given feature contributes towards prediction results. The intended system will enable patients and clinicians which directly contributes patient safety and quality of life. Future developments involve merging real-time data from continuous glucose monitoring, personalization of prediction models for individualized patients. This research establishes the foundation for an extended diabetes management tool that closes the gap between data science and medicine.