overview.
DiabetesGuard AI predicts diabetes risk using an XGBoost classifier tuned to 97.78% ROC-AUC, then explains every prediction with SHAP so clinicians can see which features drove the score. A retrieval-augmented generation layer backed by FAISS and DeepSeek V3 lets users ask follow-up questions and get answers grounded in medical literature rather than hallucinated ones. The frontend is a Streamlit dashboard built with TensorFlow/Keras for the deep learning components.
key features.
- 97.78% ROC-AUC diabetes risk classifier
- SHAP-based explainability for every prediction
- RAG-grounded Q&A over medical literature via FAISS + DeepSeek V3