A role-based healthcare decision-support prototype that uses an offline-trained SVM model to predict diabetes risk, providing dashboards, patient history, feedback, and downloadable PDF reports through a Flask web interface.
Course: CS619 — Final Year Project
Student: Hanzala Iftikhar
This system is a prototype decision-support tool built for educational purposes only.
It does not replace professional medical advice, diagnosis, or treatment.
Always consult a qualified healthcare provider for any medical concerns.
- 🩺 Diabetes Risk Prediction Web Application
- Diabetes Risk Prediction using a trained SVM (Support Vector Machine) model
- Three-level risk output: Low / Medium / High with tailored health recommendations
- Downloadable PDF reports for each prediction
- Patient Dashboard with risk distribution charts (Chart.js)
- Prediction History — patients can view all their past predictions
- Doctor Dashboard — view all patients' records across the system
- Admin Panel — manage users, add doctor accounts, view system stats
- Diabetes Education page with risk factors and prevention guidelines
- Feedback System with star rating and comments
- Role-based access control — each role sees only what it should
- Password hashing using Werkzeug's secure hash functions
| Role | How Created | Access |
|---|---|---|
| Patient | Self-registration at /register |
Prediction form, their own dashboard, history, PDF reports, education, feedback |
| Doctor | Created by Admin only | Read-only view of all patients' prediction records and risk statistics |
| Admin | Created via create_admin.py script |
User management, system stats, ability to add doctor accounts |
| Layer | Technology |
|---|---|
| Backend | Python 3.x, Flask 3.x |
| ML Model | scikit-learn (SVM), joblib |
| Data Processing | NumPy |
| Database | SQLite (via Python's built-in sqlite3) |
| PDF Generation | fpdf2 |
| Password Security | Werkzeug |
| Frontend | HTML5, CSS3, JavaScript |
| Charts | Chart.js (CDN) |
| Environment Config | python-dotenv |
The machine learning model was trained in Diabetes_FYP.ipynb using the Pima Indians Diabetes Dataset.
Pipeline:
- Dataset: 768 samples, originally 9 columns
- Feature Dropping:
Skinandtestcolumns removed (too many zero values) - Zero-Value Imputation:
pres,plas,masszero values replaced with column medians - Final Features (6):
preg,plas,pres,mass,pedi,age - Train/Test Split: 70% train, 30% test (
random_state=42) - SMOTE Balancing: Applied on training set only to address class imbalance
- Feature Scaling:
StandardScalerfitted on SMOTE-balanced training data - Models Compared:
| Model | Accuracy | Precision | Recall | F1 Score |
|---|---|---|---|---|
| Logistic Regression | 70.1% | 55.9% | 65.0% | 60.1% |
| Decision Tree | 69.7% | 54.9% | 70.0% | 61.5% |
| SVM ✅ (Selected) | 71.4% | 56.5% | 76.3% | 64.9% |
| Neural Network | 74.9% | 62.5% | 68.8% | 65.5% |
- Why SVM? Recall was prioritised for medical risk screening — it is more important to correctly identify diabetic cases (avoid false negatives) than to maximise accuracy.
- Saved Artifacts:
diabetes_model.pkl(SVM model) andscaler.pkl(fitted StandardScaler)
The Flask app (app.py) loads the saved model and scaler at startup using project-safe paths:
model = joblib.load(MODEL_PATH)
scaler = joblib.load(SCALER_PATH)On prediction, the 6 input values are scaled using the same scaler, then passed to the model:
input_data = np.array([[preg, plas, pres, mass, pedi, age]])
input_scaled = scaler.transform(input_data)
result = model.predict(input_scaled)[0]preg → plas → pres → mass → pedi → age. Changing this order will produce incorrect predictions without any error.
diabetes-risk-predictor/
│
├── app.py # Main Flask application — all routes and logic
├── create_admin.py # One-time script to create the admin account
├── requirements.txt # Python package dependencies
├── .gitignore # Files excluded from version control
├── .env # Secret key config — NOT committed to Git
├── .env.example # Template showing which env vars are needed
├── LICENSE # MIT License
├── README.md # This file
│
├── diabetes_model.pkl # Trained SVM model (committed — no user data)
├── scaler.pkl # Fitted StandardScaler (committed — no user data)
├── Diabetes_FYP.ipynb # Full ML training notebook
├── pima-indians-diabetes.data.csv # Original dataset
│
└── templates/
├── index.html # Patient prediction form (home page)
├── login.html # Login page
├── register.html # Patient self-registration
├── result.html # Prediction result + PDF download
├── dashboard.html # Patient dashboard with charts
├── history.html # Patient prediction history
├── doctor.html # Doctor dashboard (all patients)
├── admin.html # Admin panel (user management)
├── education.html # Diabetes education content
├── feedback.html # Star rating feedback form
└── access_denied.html # Access denied page for unauthorized role access
Note:
diabetes.dbis excluded from Git (see.gitignore). A fresh database is created automatically when you first run the app.
- Python 3.12 recommended
- pip
git clone https://github.com/your-username/diabetes-risk-predictor.git
cd diabetes-risk-predictor# Windows
python -m venv venv
venv\Scripts\activate
# macOS / Linux
python3 -m venv venv
source venv/bin/activatepip install -r requirements.txtCreate a .env file in the project root (copy from the example):
cp .env.example .envThen open .env and set a secure secret key:
SECRET_KEY=your-long-random-secret-key-here
You can generate a good secret key with:
python -c "import secrets; print(secrets.token_hex(32))"Run this script once to create the admin user:
python create_admin.pyThis creates the database (
diabetes.db) automatically if it does not exist.
Change the credentials increate_admin.pybefore running — do not use the defaults in production.
python app.pyOpen your browser and go to: http://127.0.0.1:5000
⚠️ These are sample placeholder credentials for documentation purposes only.
Do NOT use these in a real deployed environment.
Create your own accounts usingcreate_admin.pyand the/registerpage.
| Role | Password | |
|---|---|---|
| Admin | admin@gmail.com | admin123 or the password set in create_admin.py |
| Doctor | doctor@example.com | (set via admin panel) |
| Patient | patient@example.com | (set via /register) |
| Page | Description |
|---|---|
| Login | Role-based login screen |
| Prediction Form | 6-field input form for health parameters |
| Result Page | Risk level display with recommendations |
| Patient Dashboard | Charts + recent predictions |
| Doctor Dashboard | All patients' records overview |
| Admin Panel | User management + model performance table |
- Never commit
diabetes.db— it contains user account data. It is listed in.gitignore. - Never commit
.env— it contains the Flask secret key. It is listed in.gitignore. - Never commit real credentials in
create_admin.py— update the defaults before use. - The
diabetes__env/virtual environment folder should never be committed — it is OS-specific and very large. - Passwords are stored as salted hashes using Werkzeug — plaintext passwords are never saved.
- This is a local development / educational prototype. Before any public deployment, additional hardening is required (CSRF protection, HTTPS, rate limiting, etc.).
- Add CSRF protection for forms
- Add CSV report export
- Add admin-based model retraining in a future version
- Add pagination for large prediction records
- Deploy the application to a cloud platform