MediCare — Diabetes & Depression ML Predictions
Django health app wrapping production-style sklearn pipelines for diabetes risk and depression risk assessment with validation and explainability.
Overview
MediCare packages trained ML artifacts and prediction APIs inside a Django product shell, including documented diabetes RandomForest pipelines and a high-AUC depression risk model with engineered features.
Problem
Health-risk ML work often stops at notebooks. MediCare focuses on deployable, validated prediction paths.
Solution
Train/evaluate sklearn pipelines with medical-style input validation, persist artifacts, and serve structured JSON predictions through Django (with standalone Flask API modules available).
Architecture
Features → preprocessing/scaler → trained model → risk classification + feature importance → Django/Flask prediction endpoints.
Implementation
Diabetes: feature engineering, SMOTE, GridSearchCV, artifact persistence. Depression: 27 features (13 base + 14 engineered) with documented Accuracy 85.33%, AUC-ROC 0.9248. Django + django-allauth shell.
Results
Validated prediction engines ready for demo/integration with clear metrics and structured outputs.
Features
- Diabetes risk prediction with strict validation and risk levels
- Depression risk model with engineered features and strong AUC
- Model artifact persistence (model/scaler/metrics)
- Health and model-info endpoints
- Django app shell with optional Google OAuth
Outcomes
- Depression model AUC-ROC 0.9248
- Depression Accuracy 85.33% / F1 87.17%
- Diabetes pipeline with explainability and structured JSON output