MohamedOuakki/Algerian-Real-Estate-Predictor

ML model to predict real estate prices in Algeria using Ouedkniss data # 🏠 Algerian Real Estate Price Predictor A machine learning system that predicts real estate prices in Algeria, built end-to-end from raw data collection to a deployed, explainable web app. ## Live Demo Try the app here ## Overview This project predicts property prices across Algeria using listings data, and explains *why* the model makes each prediction using SHAP. It covers the full ML lifecycle: data collection, cleaning, feature engineering, model comparison, explainability, and deployment. ## Key Results | Model | MAE (DA) | R² | |---|---|---| | Linear Regression | 1,065,916 | 0.508 | | Random Forest | 802,865 | 0.614 | | Gradient Boosting | 905,697 | 0.607 | | **XGBoost (best)** | **809,961** | **0.624** | Trained on 15,532 cleaned listings after removing outliers and non-Algerian entries from a raw dataset of 64,099 records. ## SHAP Feature Importance Listing type (sale vs. rental) and location in Algiers emerged as the dominant price drivers — consistent with market intuition, while also revealing that the absence of structural features (surface area, floor number) limits the model's explanatory power within a given city. ## Tech Stack - **Data**: pandas, Selenium (initial scraping attempt), Kaggle dataset - **ML**: scikit-learn, XGBoost - **Explainability**: SHAP - **App**: Streamlit - **Language**: Python ## Project Structure ``` ├── data/ │ ├── raw/ # original data │ └── processed/ # cleaned, feature-engineered data ├── src/ │ ├── scraper.py # Selenium-based scraper (Ouedkniss) │ ├── preprocess.py # cleaning & feature engineering │ └── train.py # model training, comparison & SHAP ├── models/ # saved model, explainer, feature names ├── reports/ # SHAP visualizations ├── app.py # Streamlit web app └── requirements.txt ``` ## What I Learned - Built a Selenium-based scraper and diagnosed why it failed against Ouedkn …