Predicting Type 1 Diabetes Progression Using Deep Learning on Continuous Glucose Monitoring Data

This study presents a predictive framework for assessing the progression of Type 1 Diabetes by leveraging key glycemic variability metrics, Dynamic Stress Factor, Mean Amplitude of Glycemic Excursions, Mean of Daily Differences, Continuous Overall Net Glycemic Action alongside patient age. Two modeling approaches are explored: a machine learning model using a Random Forest Classifier and a deep learning model based on Recurrent Neural Networks. Patients are classified into three clinically significant categories: Stage 2 Type 1 Diabetes, Stage 1 Type 1 Diabetes, and Low Risk. After addressing class imbalance with Synthetic Minority Oversampling Technique, the Random Forest model achieved 99.6% accuracy, while the RNN model reached 100% accuracy on the test dataset. Feature importance analysis revealed that Dynamic Stress Factor and Continuous Overall Net Glycemic Action were the most predictive features, emphasizing the critical role of glycemic volatility in early diabetes detection. In contrast, static parameters like age and mean glucose showed minimal contribution. These findings underscore the effectiveness of deep learning models in capturing temporal glucose patterns and support the clinical utility of time-series-based metrics for personalized diabetes management. The proposed framework offers a promising tool for improving early diagnosis, guiding intervention strategies, and enhancing clinical decision support in Type 1 Diabetes care.