GLUCOTWIN: AN ARTIFICIAL INTELLIGENCE–DRIVEN INSULIN DOSAGE PREDICTION SYSTEM

Abstract

Diabetes management in children requires continuous monitoring and accurate insulin dosage adjustments based on multiple physiological and dietary factors. Incorrect insulin administration can lead to serious health complications, particularly when caregivers lack the necessary medical expertise to make informed decisions. This study presents GLUCOTWIN, an AI-driven healthcare application that assists with insulin dosage prediction for children aged 3–12 years by integrating machine learning into a user-friendly mobile platform. The proposed system utilizes health-related parameters, including blood glucose level, blood pressure, body temperature, Body Mass Index (BMI), meal timing, and carbohydrate intake, to predict personalized insulin dosage recommendations. A role-based architecture was developed comprising Admin, Guardian, and Doctor modules, enabling secure data management, medical supervision, and real-time decision support. The mobile application was developed using Flutter, while Firebase was employed as the backend database for storing patient records, prediction history, and user information. Multiple machine learning regression algorithms were trained and evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R² Score. Among the evaluated models, the Gradient Boosting Regressor achieved the highest predictive performance with an MAE of 0.2921, RMSE of 0.3736, and R² score of 0.9570, demonstrating high accuracy and reliability in insulin prediction. The trained model was successfully deployed and integrated into the application to provide real-time insulin dosage recommendations, which can be reviewed and validated by healthcare professionals. The results indicate that GLUCOTWIN has the potential to enhance diabetes management, support caregivers in making informed decisions, and contribute to improved healthcare outcomes for paediatric diabetes patients