🧠 Real Estate Price Prediction with XGBoost

This project aims to build a machine learning pipeline to train and evaluate a model using XGBoost on a real estate dataset provided in CSV format. It is designed with modular components to handle data preprocessing, encoding, training, evaluation, and prediction.

🌐 Live Demo on render: https://immo-predict.onrender.com

🌐 Live Demo on railway: https://immo-eliza-predict.up.railway.app

🌐 Live Demo on HuggingFace: https://huggingface.co/spaces/Fillinger66/immo-eliza-demo

---

📁 Project Structure

├── data/ # Raw input data (CSV files)
│ └── *.csv
│
├── lib/ # Core library code
│ ├── encoders/
│ │ └── TopKEncoder.py # Custom encoder for top-K categories
│ │
│ ├── model/
│ │ └── XGBoostModel.py # Wrapper for XGBoost model
│ │
│ ├── DataCleaner.py # Optional: cleaning/preprocessing logic
│ ├── DataClustering.py # Optional: clustering operations (e.g., KMeans)
│ ├── DataManager.py # File I/O and data manipulation
│ ├── DataMetrics.py # Model evaluation metrics (R², MAE, RMSE)
│ ├── DataPipeline.py # ML preprocessing pipeline (scikit-learn style)
│
├── model/ # Saved model files
│ └── *.model
│
├── pipeline/ # Optional: pipeline configurations or artifacts
│ └── *?pipeline # (Clarify what's inside)
│
├── run.py # Main execution script
├── requirements.txt # Python dependencies
└── README.md # Project documentation\

Features

• DataPipeline: Used to create the pipeline.
• DataManager: Used to interact with files (load CSV, merge DataFrame columns, etc.).
• DataMetric: Used to get metrics like R², MAE, RMSE, etc.
• XGBoostModel: Used to create, train, and predict using an XGBoost model.
• TopKEncoder: Used as a pipeline encoder to get the top K categories to reduce the number of columns.
• Run script: Used to train, predict, etc., and uses DataPipeline, DataManager, and XGBoostModel.

---

🎯 Purpose

The goal is to create a reproducible pipeline to:

1. Preprocess and encode real estate dataset features
2. Train an XGBoost model
3. Predict on unseen data
4. Evaluate using regression metrics like R², MAE, and RMSE

---

🧩 Components

DataPipeline

• Builds a preprocessing pipeline using scikit-learn and custom encoders.
• Handles:
  • Missing value imputation
  • Label encoding
  • Boolean transformation
  • Train/test split
• Utilizes TopKEncoder for categorical column compression.

---

DateManager

• Responsible for file operations:
  • Loading CSV files
  • Merging column values into new derived features
• Acts as a utility class to manage I/O operations.

---

DataMetric

• Calculates evaluation metrics:
  • R² (R-squared)
  • MAE (Mean Absolute Error)
  • RMSE (Root Mean Square Error)
• Used during validation and model performance analysis.

---

XGBoostModel

• Wraps XGBoost regressor for training and inference.
• Supports:
  • Custom hyperparameters
  • Model saving/loading
  • Feature importance extraction

---

TopKEncoder

• Custom categorical encoder to reduce one-hot encoding dimensionality.
• Retains only top-K frequent categories for each column.
• Reduces feature space and risk of overfitting.

---

run.py

• Main script to execute the end-to-end ML pipeline:
  • Loads data using DateManager
  • Preprocesses via DataPipeline
  • Trains model using XGBoostModel
  • Evaluates performance with DataMetric

---

🛠 Requirements

pgeocode
pandas
numpy
scikit-learn
xgboost
tensorflow
geopy
matplotlib
plotly

🚀 How to Run

# Install dependencies
pip install -r requirements.txt

# Run the pipeline
python src/run.py

Author

Alexandre Kavadias