🩺 Diabetes Risk Predictor (SDG 3 – Good Health & Well-being)

This project is a simple offline health risk prediction tool that estimates the risk of type 2 diabetes using basic health parameters. It uses a Logistic Regression model trained on the Pima Indians Diabetes Dataset and a Streamlit web app for interaction.

The app supports Sustainable Development Goal 3 (SDG 3): Good Health & Well-being by:

• Raising awareness about diabetes risk
• Encouraging early screening and preventive behaviour
• Demonstrating how low-cost, offline AI tools can support health decision-making

⚠️ Disclaimer: This project is for education and demonstration only.
It is NOT a medical device and must not be used for real diagnosis or treatment.

---

PitchDeck

🖼️ Screenshots

1. Risk Prediction Tab

---

2. About & SDG 3 Tab

---

3. Usage Stats Tab

🗂️ Project Structure

diabetes_project/
  ├── screenshots/
  │   ├── prediction_tab.png
  │   ├── about_tab.png
  │   └── usage_stats_tab.png
  ├── app.py
  ├── train_model.py
  ├── diabetes_model.pkl
  ├── diabetes.csv
  └── README.md

---

## 💻 1. Requirements

You need:

* **Python 3.8+**
* Recommended: a virtual environment (optional but clean)

Install the required Python packages:

```bash
pip install streamlit scikit-learn pandas numpy

---

📥 2. Dataset

This project uses the Pima Indians Diabetes Dataset. Make sure you have a file named diabetes.csv in the project folder with columns like:

• Pregnancies
• Glucose
• BloodPressure
• SkinThickness
• Insulin
• BMI
• DiabetesPedigreeFunction
• Age
• Outcome (0 = no diabetes, 1 = diabetes)

If you downloaded the dataset from Kaggle or GitHub, just rename the CSV to diabetes.csv and place it in the same folder as train_model.py.

---

🧠 3. Training the Model

Run the training script to:

• Load and clean the dataset
• Split it into train/test sets
• Train a Logistic Regression model
• Save the model and scaler into diabetes_model.pkl

From the project folder, run:

python train_model.py

You should see output including:

• Accuracy on the test set
• Classification report
• Confusion matrix
• A message like: Model saved to diabetes_model.pkl

After this, you should have:

diabetes_model.pkl

in your folder.

---

🌐 4. Running the Streamlit App

Once the model file (diabetes_model.pkl) exists, start the app with:

streamlit run app.py

This will open the app in your browser at:

http://localhost:8501

The app has three main tabs:

🔮 Risk Prediction

• Enter your health parameters in the sidebar:

  • Pregnancies
  • Glucose (mg/dL)
  • Blood Pressure (mm Hg)
  • Skin Thickness (mm)
  • Insulin (mu U/ml)
  • BMI (kg/m²)
  • Diabetes Pedigree Function
  • Age (years)

• Click “Predict Risk”.

• The app will display:

  • A prediction (low diabetes risk vs diabetes risk)
  • Estimated probability of diabetes (0–1)
  • A risk level: Low / Medium / High
  • General, non-medical health tips and a disclaimer

All predictions are made locally and offline using the saved model.

ℹ️ About & SDG 3

This tab explains:

• The goal of the project
• Connection to SDG 3: Good Health & Well-being
• Basic information about the dataset and model
• Important disclaimer: the app is not a diagnostic tool

This is useful for reports, presentations, or anyone viewing the project for the first time.

📈 Usage Stats

Each time the user makes a prediction, the app logs it to usage_log.csv (created automatically). The log does not contain any names or IDs, just:

• Timestamp
• Input health parameters
• Predicted class
• Probability
• Risk level (Low/Medium/High)

The Usage Stats tab shows:

• Total number of predictions
• Average predicted probability of diabetes
• A bar chart of risk levels (Low/Medium/High)
• A table showing the last 10 predictions

If usage_log.csv does not exist yet, the tab will show a message asking you to make some predictions first.

---

🧪 5. How It Works (Technical Overview)

1. Data Preprocessing

   • Load diabetes.csv using pandas.

   • Replace unrealistic zeros in:

     • Glucose, BloodPressure, SkinThickness, Insulin, BMI

   • Impute these with the median of each column.

   • Separate features (X) and target (y = Outcome).

2. Train–Test Split

   • Split data into 80% training and 20% testing.
   • Use stratified sampling to maintain class balance.

3. Scaling

   • Use StandardScaler to scale the features.
   • Fit on training data, transform both train and test, and later user input.

4. Model

   • Use Logistic Regression (sklearn.linear_model.LogisticRegression).
   • Train on the scaled training data.
   • Evaluate on the test set (accuracy, precision, recall, etc.).

5. Saving Model

   • Save a Python dictionary containing:

     • model – the trained Logistic Regression model
     • scaler – the fitted StandardScaler
     • feature_names – list of feature names

   • Use pickle.dump() to store it in diabetes_model.pkl.

6. Streamlit App

   • Load diabetes_model.pkl at startup.
   • Collect user input via Streamlit widgets in the sidebar.
   • Convert input to a NumPy array, scale with the loaded scaler.
   • Call model.predict() and model.predict_proba() to generate predictions.
   • Display results + risk levels + health tips.
   • Log each prediction to usage_log.csv for stats.

---

📊 6. Example Output / Metrics

When you run train_model.py, you’ll get evaluation metrics on the test set, such as:

• Test accuracy (e.g., ~0.75–0.80)
• Precision and recall for both classes
• Confusion matrix

You can copy these results into your report or presentation.

---

🌍 7. SDG 3 – Good Health & Well-being

This project demonstrates how AI and simple web tools can support:

• Awareness of non-communicable diseases (like diabetes)
• Early screening mindset, by prompting users to think about risk
• Low-cost, offline tools that can be run anywhere with a laptop

Even though this tool is not clinically validated, it illustrates important ideas behind using data and machine learning to support public health, linking directly to SDG 3.

---

⚠️ 8. Limitations

• Dataset is limited to a specific population (Pima Indian women).
• Only 8 numerical features are used. Real diagnosis requires more data and clinical judgement.
• Logistic Regression is simple and may not be the most accurate possible model.
• The model is not clinically validated and must not be used for real patient care.

---

🔮 9. Possible Future Improvements (Optional)

These are ideas only – not implemented in this version:

• Try and compare other models (e.g., Random Forest, Gradient Boosting).
• Add more visual explanations for how each feature affects the prediction.
• Extend the app to predict risk for other diseases (e.g., heart disease).
• Improve content and explanations for different audiences.
• Add support for multiple languages.

---

✅ 10. How to Run (Quick Steps)

1. Clone or copy this project folder.

2. Place diabetes.csv in the project folder.

3. Install dependencies:

   pip install streamlit scikit-learn pandas numpy

4. Train the model:

   python train_model.py

5. Run the app:

   streamlit run app.py

6. Open your browser at http://localhost:8501 and start exploring.