Car Price Prediction API

AI-powered car price prediction microservice using LightGBM machine learning. This service is part of the autoanaliza.pl platform.

🚀 Live Demo

Try it live at www.autoanaliza.pl - create a free account to test the AI price predictions in action!

Overview

This microservice provides real-time car price predictions based on vehicle specifications. It uses a LightGBM gradient boosting model trained on hundreds of thousands of car listings from the Polish automotive market.

Key Features

• Fast predictions: < 10ms response time per vehicle
• Accuracy: ~8,277 PLN MAE on test set
• Handles missing data: Works with incomplete vehicle information
• RESTful API: Easy integration via FastAPI
• Automatic feature engineering: Creates derived features for better predictions

Tech Stack

• Python 3.12
• FastAPI - Modern web framework for building APIs
• LightGBM - Gradient boosting framework for machine learnings
• PostgreSQL - Database for car listings
• scikit-learn - Machine learning utilities
• pandas & numpy - Data processing

How It Works

The model uses 18 features to predict car prices:

Base Features (13)

• Make, model, year, body type, fuel type
• Engine displacement (cc), engine power (HP)
• Transmission, drive type, mileage
• Seller type, damage status, color, steering side

Engineered Features (5)

• Car age (calculated from year)
• Mileage per year (mileage / age)
• Power-to-displacement ratio
• Average price for make
• Average price for model

The model automatically normalizes text inputs and handles missing values using LightGBM's native capabilities.

API Endpoints

GET /

Health check and API information

GET /health

Detailed health status with model statistics

POST /predict

Predict car price based on vehicle specifications

Request body:

{
  "make": "audi",
  "model": "a5",
  "year": 2013,
  "body_type": "coupe",
  "fuel": "benzyna",
  "engine_cc": 1984,
  "engine_power": 211,
  "transmission": "manualna",
  "drive": "awd",
  "mileage": 150000,
  "seller_type": "private",
  "is_damaged": false,
  "color": "niebieski",
  "right_hand": false
}

Response:

{
  "predicted_price": 45000.0,
  "confidence_range": {
    "min": 38250.0,
    "max": 51750.0
  },
  "input_data": { ... }
}

Installation & Setup

Prerequisites

• Python 3.10+
• PostgreSQL database with car listings

1. Clone the repository

git clone https://github.com/yourusername/estimate-car-value-ai.git
cd estimate-car-value-ai

2. Install dependencies

pip install -r requirements.txt

3. Configure environment

Create a .env file:

DBNAME=your_database
USER=your_user
PASSWORD=your_password
HOST=your_host
PORT=5432

4. Train the model (optional)

python car_price_model.py

This will:

• Load data from PostgreSQL
• Engineer features
• Train the LightGBM model
• Save the model to car_price_model.pkl

5. Run the API

python api.py

The API will be available at http://localhost:8000

Usage Examples

Python

import requests

response = requests.post('http://localhost:8000/predict', json={
    'make': 'audi',
    'model': 'a5',
    'year': 2013,
    'fuel': 'benzyna',
    'mileage': 150000
})

result = response.json()
print(f"Predicted price: {result['predicted_price']:,.0f} PLN")
print(f"Range: {result['confidence_range']['min']:,.0f} - {result['confidence_range']['max']:,.0f} PLN")

cURL

curl -X POST "http://localhost:8000/predict" \
  -H "Content-Type: application/json" \
  -d '{
    "make": "audi",
    "model": "a5",
    "year": 2013,
    "fuel": "benzyna",
    "mileage": 150000
  }'

Model Performance

Trained on ~636,000 car listings:

• MAE (Mean Absolute Error): 8,277 PLN
• R² Score: 0.9
• Training time: 2 minutes
• Prediction time: < 10ms per vehicle

Project Structure

├── api.py                  # FastAPI application
├── car_price_model.py      # Model training and prediction logic
├── fetch_data.py           # Database data fetching utilities
├── analyze_model.py        # Model analysis tools
├── test_predictions.py     # Testing utilities
├── requirements.txt        # Python dependencies
├── Dockerfile             # Docker configuration
└── car_price_model.pkl    # Trained model (generated)

Docker Deployment

docker build -t car-price-api .
docker run -p 8000:8000 --env-file .env car-price-api

License

MIT License - feel free to use this project for your own purposes.

About autoanaliza.pl

This microservice is part of autoanaliza.pl, a comprehensive platform for analyzing the Polish automotive market. Create a free account to access:

• AI-powered price predictions
• Market analysis and trends
• Vehicle history insights
• And more!

---

Note: This model is optimized for the Polish automotive market and works best with vehicles from 1990-2025 with prices between 1,000-3,000,000 PLN.