Car Price Prediction Project 🚗

Table of Contents

1. Project Overview

2. Tech Stack

3. Architecture & Workflow

4. Data Ingestion

5. Data Transformation

6. Model Training & Evaluation

7. Prediction Pipeline

8. FastAPI Deployment

9. MLflow & DagsHub Integration

10. Dockerization

11. CI/CD Pipeline

12. Project Structure

13. References

Project Overview

This project predicts car prices in Pakistan using historical car data. Features include:

• Company / Brand

• Car Model

• Year of Manufacture

• Kilometers Driven

• Fuel Type (Petrol / Diesel)

The project implements a complete ML workflow including data ingestion, preprocessing, modeling, deployment, and CI/CD.

Tech Stack

• Backend & ML: Python, Pandas, NumPy, Scikit-learn, XGBoost

• API & Web: FastAPI, Jinja2 Templates, JavaScript, HTML/CSS

• Database: MySQL

• Experiment Tracking: MLflow, DagsHub

• Containerization: Docker

• CI/CD: GitHub Actions / GitLab CI

• Version Control: Git / DagsHub

Architecture & Workflow

flowchart TD
    A[MySQL Database] --> B[Data Ingestion]
    B --> C[Data Transformation & Feature Engineering]
    C --> D[Model Training & Evaluation]
    D --> E[Prediction Pipeline]
    E --> F[FastAPI Backend]
    F --> G[Frontend UI]
    D --> H[MLflow & DagsHub Tracking]
    F --> I[Docker Container]
    I --> J[CI/CD Pipeline: GitHub Actions / GitLab CI]

Data Ingestion

• Data is fetched from MySQL using Python (pymysql).

• Split data into train and test sets.

Data Transformation

Feature Engineering:

• age = 2025 - year

• One-hot encode categorical features: company, name, fuel_type

• Scale numerical features: age, kms_driven

Preprocessor object is saved for later prediction: artifacts/preprocessor.pkl

Model Training & Evaluation

Train a regression model (XGBoost / RandomForest) using the transformed dataset.

• Evaluate metrics:

• RMSE, MAE, R²

• Save trained model: artifacts/model.pkl

= Log metrics with MLflow.

Prediction Pipeline

predict_pipeline.py handles:

• Input validation using Pydantic

• Feature engineering (computing age)

• Transformation using saved preprocessor

• Prediction using saved model

Example:

pipeline = PredictionPipeline(model_path="artifacts/model.pkl",
                              preprocessor_path="artifacts/preprocessor.pkl")
prediction = pipeline.predict(input_df)

FastAPI Deployment

Endpoints:

• / → Homepage with prediction form

• /predict → POST API for predictions

• /company → GET all companies

• /name/{company_name} → GET car models per company

• Frontend Integration:

• HTML/CSS/JS form

• AJAX calls to API endpoints

• Dynamic display of predicted price

Example request payload:

{
  "name": "Civic",
  "company": "Honda",
  "year": 2018,
  "kms_driven": 45000,
  "fuel_type": "Petrol"
}

MLflow & DagsHub Integration

• Track experiments, metrics, and parameters with MLflow.

• Example:

import mlflow
mlflow.set_tracking_uri("https://dagshub.com/<username>/<repo>.mlflow")
mlflow.log_param("model", "XGBRegressor")
mlflow.log_metric("RMSE", rmse)
mlflow.sklearn.log_model(best_model, "model")

Benefits:

• Model versioning

• Experiment comparison

• Collaboration via DagsHub

Dockerization

Dockerfile:

FROM python:3.10-slim

WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .

EXPOSE 8000
CMD ["uvicorn", "src.mlproject.main:app", "--host", "0.0.0.0", "--port", "8000"]

Commands:

docker build -t car-price-predictor .
docker run -d -p 8000:8000 car-price-predictor

CI/CD Pipeline

• Automate build, test, and deployment using GitHub Actions or GitLab CI.

Sample GitHub Action workflow:

name: CI/CD

on:
  push:
    branches: [main]

jobs:
  build-deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - name: Set up Python
        uses: actions/setup-python@v4
        with:
          python-version: 3.10
      - name: Install dependencies
        run: pip install -r requirements.txt
      - name: Run tests
        run: pytest
      - name: Build Docker
        run: docker build -t username/car-price-predictor .
      - name: Push Docker
        run: docker push username/car-price-predictor

Project Structure

Car-Price-Predictor/ │

├─ src/

│ ├─ mlproject/

│ │ ├─ components/

│ │ │ ├─ data_ingestion.py

│ │ │ ├─ data_transformation.py

│ │ │ ├─ model_trainer.py

│ │ ├─ pipelines/

│ │ │ └─ prediction_pipeline.py

│ │ ├─ utils.py

│ │ ├─ logger.py

│ │ └─ main.py

│

├─ artifacts/

│ ├─ raw_data.csv

│ ├─ train.csv

│ ├─ test.csv

│ ├─ model.pkl

│ └─ preprocessor.pkl

│

├─ templates/

│ └─ index.html

├─ static/

│ └─ style.css

├─ requirements.txt ├─ Dockerfile └─ README.md

References

• FastAPI Docs

• MLflow Docs

• DagsHub Docs

• Docker Docs