MT: Bioinformatics, diabetes mellitus, deep learning, model interpretability, risk stratification, diabetes risk prediction, Therapeutics. Pharmacology, RM1-950

Diabetes is a prevalent chronic disease that poses a significant burden on individuals and healthcare systems. Early diagnosis and effective management are essential for delaying disease onset and minimizing complications. In this study, we developed a deep learning-based risk prediction model using data from the National Health and Nutrition Examination Survey (NHANES) spanning from 2011 to 2018. From an initial set of 36 variables, and through feature selection, five key features were identified through feature selection techniques to construct the model. The model demonstrated robust performance, accurately predicting diabetes risk with high precision in test data. External validation further confirmed its ability to correctly identify individuals at risk of developing diabetes. To enhance its practical application, we implemented a risk stratification system and developed a user-friendly online tool, available at Visa/, allowing easy access for users. This model provides a valuable tool for diabetes risk screening and personalized early detection.