Health Oracle — ML API

Lightweight FastAPI service that serves three pre-trained health prediction models:

• Obesity prediction
• Insomnia prediction
• Diabetes prediction

---

Repository contents

• main.py — FastAPI application and route definitions
• requirements.txt — Python dependencies
• Obesity_Prediction_Model.pkl — obesity model (default)
• Obesity_Label_Encoder.pkl — encoder(s) for obesity model (default)
• Insomnia_model.pkl — insomnia model (default)
• Insomnia_label_encoders.pkl — encoder(s) for insomnia model (default)
• diabetes_prediction_model.pkl — diabetes model (default)

---

Requirements

• Python 3.8+
• pip
• (Optional) virtualenv/venv

Install dependencies: pip install -r requirements.txt

---

Environment variables

main.py loads model file paths from environment variables with fallback defaults:

• OBESITY_MODEL — path to obesity model (default: Obesity_Prediction_Model.pkl)
• OBESITY_ENCODER — path to obesity encoder(s) (default: Obesity_Label_Encoder.pkl)
• INSOMNIA_MODEL — path to insomnia model (default: Insomnia_model.pkl)
• INSOMNIA_ENCODER — path to insomnia encoder(s) (default: Insomnia_label_encoders.pkl)
• DIABETES_MODEL — path to diabetes model (default: diabetes_prediction_model.pkl)

You can place these in a .env file for local development or configure them in your deployment environment.

---

Run locally

Start the app (FastAPI app object name is app in main.py): uvicorn main:app --reload --host 0.0.0.0 --port 8000

Health check: GET http://localhost:8000/ Response: { "message": "HealthOracle FastAPI is up!" }

---

API Endpoints

All endpoints accept JSON and return JSON with a "prediction" key.

1. POST /predict/obesity

• Input model: ObesityInput (field names and casing must match exactly) { "Gender": "Female", "Age": 29, "Height": 160.0, "Weight": 60.0, "family_history_with_overweight": "no", "FAVC": "no", "FCVC": 2.0, "NCP": 2.0, "CAEC": "Sometimes", "SMOKE": "no", "CH2O": 2.0, "SCC": "no", "FAF": 1.0, "TUE": 1.0, "CALC": "Sometimes", "MTRANS": "Walking" }

Example curl: curl -X POST http://localhost:8000/predict/obesity
-H "Content-Type: application/json"
-d '{"Gender":"Female","Age":29,"Height":160.0,"Weight":60.0,"family_history_with_overweight":"no","FAVC":"no","FCVC":2.0,"NCP":2.0,"CAEC":"Sometimes","SMOKE":"no","CH2O":2.0,"SCC":"no","FAF":1.0,"TUE":1.0,"CALC":"Sometimes","MTRANS":"Walking"}'

Example Python: import requests resp = requests.post("http://localhost:8000/predict/obesity", json={...}) print(resp.json())

Response: {"prediction": "<label_or_value>"}

Notes:

• main.py uses obesity_encoder.transform([input_dict]) before predicting — the encoder must accept a dict-like mapping for transform. If you get errors, convert the input into a pandas DataFrame with the correct column order as used during training.

2. POST /predict/insomnia

• Input model: InsomniaInput { "Gender": "Male", "Age": 40, "Occupation": "Office", "Sleep_Duration": 6.0, "Quality_of_Sleep": 2, "Physical_Activity_Level": 2, "Stress_Level": 3, "BMI_Category": "Overweight", "Blood_Pressure": "Normal", "Heart_Rate": 72, "Daily_Steps": 4000 }

Example curl: curl -X POST http://localhost:8000/predict/insomnia
-H "Content-Type: application/json"
-d '{"Gender":"Male","Age":40,"Occupation":"Office","Sleep_Duration":6.0,"Quality_of_Sleep":2,"Physical_Activity_Level":2,"Stress_Level":3,"BMI_Category":"Overweight","Blood_Pressure":"Normal","Heart_Rate":72,"Daily_Steps":4000}'

Response: {"prediction": "<label_or_value>"}

Notes:

• main.py uses insomnia_encoder.transform([input_dict]) — ensure the saved encoder expects a mapping or adjust preprocessing to match training pipeline.

3. POST /predict/diabetes

• Input model: DiabetesInput (note: fields are lowercase) { "gender": "Female", "age": 50, "occupation": "Teacher", "sleep_duration": 6.5, "quality_of_sleep": 2, "physical_activity": 1, "stress_level": 2, "bmi_category": "Obese", "blood_pressure": "High", "heart_rate": 80, "daily_steps": 2000 }

Example curl: curl -X POST http://localhost:8000/predict/diabetes
-H "Content-Type: application/json"
-d '{"gender":"Female","age":50,"occupation":"Teacher","sleep_duration":6.5,"quality_of_sleep":2,"physical_activity":1,"stress_level":2,"bmi_category":"Obese","blood_pressure":"High","heart_rate":80,"daily_steps":2000}'

Response: {"prediction": "<label_or_value>"}

Notes:

• main.py creates a list from input.dict().values() and passes it to diabetes_model.predict([input_list]). The order of values in input.dict().values() is insertion order based on the Pydantic model definition — ensure the model expects the same feature order. If results look wrong, explicitly reorder fields into the expected model input order before predicting.

---

Troubleshooting & tips

• If you get pickling/unpickling errors, confirm the Python and library versions used during serialization match your runtime.
• Large model files (~60 MB) are included. For deployments, consider downloading models from object storage at container startup or using Git LFS.
• If encoder.transform(...) raises an error, inspect how encoders were saved (scikit-learn encoders typically expect arrays or DataFrame input) and adapt preprocessing accordingly.
• Validate incoming data strictly (add Pydantic field constraints or additional validation if needed).

---

Testing

• Use curl or Postman for quick testing.
• Add unit tests for:
  • Pydantic validation
  • Prediction pipeline (preprocessing → model)
  • Model loading at startup
  • Error handling for bad input

---

Deployment recommendations

• Serve with a production ASGI server: uvicorn + gunicorn (for FastAPI) behind a reverse proxy (nginx).
• Set model file environment variables in the deployment environment or download model artifacts at container startup.
• Lock dependency versions in requirements.txt.

---

Contributing

• Fork, create a branch, add tests, and open a pull request.
• Include a description of changes and how to test them locally.