Development of a Machine Learning-Based Interface for Insulin Dependency Prediction Using Clinical Data

This study presents the development and evaluation of machine learning models to predict insulin dependency in diabetic patients using clinical and demographic data. Utilizing a dataset comprising variables such as age, gender, BMI, HbA1c, fasting and postprandial blood sugar levels, smoking and alcohol status, and diabetes duration, we trained six models: Random Forest, Logistic Regression, XGBoost, LightGBM, a Voting Ensemble, and an Averaged Model (Random Forest + LightGBM). The models were assessed using accuracy, AUC, precision, recall, and F1-score. The LightGBM model and ensemble methods achieved the highest performance, each with an accuracy of 90% and an AUC of 0.9341, demonstrating strong predictive ability for both insulin-dependent and non-insulin-dependent groups. Feature importance analysis revealed HbA1c, duration of diabetes, and glucose levels as critical predictors. The most effective model was deployed as an interactive web interface using Gradio on Hugging Face Spaces. Our findings suggest that machine learning, particularly ensemble approaches, can provide valuable tools for early prediction of insulin needs in diabetic patients, supporting clinical decision-making and personalized care strategies.