SmartDiabX: An End-to-End AI System for Predicting Undiagnosed Type II Diabetes

Abstract

The increasing global prevalence of Type II diabetes highlights the need for accessible tools that support early detection and preventive care. While many existing artificial intelligence (AI) solutions focus on managing individuals already diagnosed with diabetes, fewer systems address the early identification of undiagnosed cases. This study presents SmartDiabX an end-to-end AI-powered application designed to predict the likelihood of Type II diabetes using user-provided health information. Multiple machine learning models, including K-Nearest Neighbors and XGBoost, were evaluated using datasets derived from the Centers for Disease Control and Prevention's Behavioral Risk Factor Surveillance System (BRFSS) and the National Health and Nutrition Examination Survey (NHANES). Initial experiments with BRFSS-based datasets demonstrated limited performance due to class imbalance and weak predictive features. By transitioning to the NHANES dataset, which includes clinically relevant indicators such as blood glucose levels, and optimizing model training with XGBoost, the proposed approach achieved an accuracy of 91% and an F1-score of 0.87. The final model was integrated into a full-stack web application utilizing React, FastAPI, and MongoDB, enabling real-time risk prediction, confidence scoring, personalized lifestyle recommendations, and longitudinal health trend visualization. The results demonstrate the potential of AI-driven systems to enhance early diabetes awareness and support preventive healthcare outside traditional clinical environments.

Ähnliche Arbeiten