Drug Recommendation System with FastAPI

A compact machine learning project that trains a drug-prescription prediction model from patient health metrics and exposes the model through a FastAPI endpoint.

What it does

• Trains a decision tree regression pipeline on the drug200.csv dataset.
• Encodes categorical patient features and predicts the most likely prescribed drug.
• Serves inference requests through a REST-style FastAPI endpoint.
• Returns a clear JSON response with input values and the recommended drug.

Key files

• drug200.csv — dataset containing patient features and prescribed drug labels.
• ml_drug_withAPI.py — training script that builds the preprocessing + model pipeline and exports two artifacts:
  • drug_pipeline_model.pkl
  • label_encoder.pkl
• main.py — FastAPI application that loads the saved artifacts and exposes a path-based inference route.

How to use

1. Install dependencies

pip install fastapi uvicorn pandas scikit-learn joblib

2. Train the model (if artifacts are not already present)

python ml_drug_withAPI.py

3. Start the API server

uvicorn main:app --reload

4. Call the prediction endpoint

Open in browser or use curl:

http://127.0.0.1:8000/docs

Example response

{
  "status": "Success",
  "extracted_path_parameters": {
    "Age": 45,
    "Sex": "M",
    "BP": "HIGH",
    "Cholesterol": "NORMAL",
    "Na_to_K": 18.5
  },
  "recommended_drug": "DrugY"
}

Replace DrugY with the actual model recommendation from your dataset.

Notes

• The API uses path parameters for inference input.
• The model expects Sex, BP, and Cholesterol values in uppercase format.
• Na_to_K must be a positive float.

Project goal

This repository demonstrates how to connect a trained Scikit-learn pipeline to a lightweight web API, making drug-prescription predictions available for real-time use.