CKD Predictor API

This repository provides a FastAPI-based web service for predicting Chronic Kidney Disease (CKD) using a machine learning model. The service is designed for easy deployment and integration into data science workflows or clinical decision support systems.

Features

• REST API for CKD prediction using a pre-trained machine learning model
• /predict endpoint accepts tabular data and returns predictions and probabilities
• /health endpoint for service health checks
• /schema endpoint to inspect the expected feature order
• Automatic input normalization (handles column name formatting and missing features)
• Configurable model path and prediction threshold via environment variables
• Sample model artifacts and configuration files included in the artifacts/ directory

How It Works

1. Model Loading: On startup, the API loads a pre-trained model (default: artifacts/model.joblib). The expected feature order is inferred from the model or from metadata/config files.
2. Prediction: The /predict endpoint accepts a JSON payload with a list of feature dictionaries. Input columns are normalized, missing features are filled with default values, and the model returns predictions and probabilities.
3. Health and Schema: The /health endpoint returns service status and expected feature count. The /schema endpoint returns the expected feature order for debugging and integration.

API Endpoints

• GET /health — Returns service status and expected feature count
• GET /schema — Returns the expected feature order
• POST /predict — Accepts a JSON payload with rows of features, returns predictions and probabilities

Example Usage

POST /predict
{
  "rows": [
    {"age": 50, "blood_pressure": 80, ...},
    {"age": 65, "blood_pressure": 90, ...}
  ]
}

Response:

{
  "pred": [0, 1],
  "proba": [0.12, 0.87],
  "threshold": 0.5,
  "received_columns": [...],
  "served_order": [...]
}

Configuration

• MODEL_PATH: Path to the model file (default: artifacts/model.joblib)
• DEFAULT_THRESHOLD: Default probability threshold for binary classification (default: 0.50)

Requirements

• Python 3.8+
• See requirements.txt for dependencies

Running Locally

1. Install dependencies:

   pip install -r requirements.txt

2. Start the API:

   uvicorn app:app --reload

Artifacts

The artifacts/ directory contains sample model files, configuration, and metadata used for inference and schema validation.

License

This project is provided for educational and research purposes. See LICENSE for details if present.