BlackMoon681/tunisia-car-price-mlops

# Used Car Price Prediction – Tunisian Market (Automobile.tn) **End-to-End MLOps Project** Predicting fair market prices for used cars in Tunisia using real-time data scraped from automobile.tn ## 🎯 Problem Description (Business & ML Objective) The Tunisian used car market suffers from **lack of transparency**: - Sellers frequently overprice vehicles due to limited visibility into real market values - Buyers have no reliable way to assess whether a listed price is fair - No up-to-date, data-driven tool exists that accounts for local specifics (governorate, fuel type, mileage in km, year, gearbox, etc.) **This project solves** the problem by delivering: 1. Automated scraping of current listings from automobile.tn 2. Structured storage in a MySQL star schema 3. A high-accuracy regression model (LightGBM) to predict price in TND 4. A production-ready REST API for real-time price predictions 5. Full MLOps stack: experiment tracking (MLflow), orchestration (Prefect), monitoring (Evidently), containerization (Docker), cloud deployment (Render) **Target users**: - Private buyers → instantly check if a car is fairly priced - Sellers & dealers → set competitive, data-informed prices - Market analysts → understand key price drivers in Tunisia **Success metrics**: - Model performance: RMSE 0.85 (on cleaned data) - End-to-end reproducibility & observability via MLOps tools ## 🏗 Project Architecture (End-to-End MLOps Pipeline) [Scraping (Selenium + BS4)] → car_listings.json ↓ [ETL → MySQL star schema] → car_data database ↓ [Cleaning + Feature Engineering] ↓ [Model Training (LightGBM + GridSearchCV + SHAP)] → car_price_model.joblib ↓ [Experiment tracking & registry (MLflow)] ↓ [FAISS semantic index (similarity search)] ↓ [Flask REST API (/predict endpoint)] ↓ [Monitoring (Evidently AI – drift detection + alerts)] ↓ [Orchestration (Prefect 2 – fully deployed local workflow)] ↓ [Containerized deployment (Docker + Render.com)] text## 🚀 Quick Start (Local) ### Prerequisites …