Customizable Real Estate Market Forecasting Tool

Project Overview

This project is a customizable real estate market forecasting tool that uses SARIMA (Seasonal AutoRegressive Integrated Moving Average) models to predict Zillow Home Value Index (ZHVI) trends at different geographical levels, such as states and cities. This tool already includes ready-to-use models with predictions for the next 11 months as of 10/06/2024 for 10 states including New York, California and Texas, the dataset is updated monthly, reducing maintenance.

Key Features

• Customizable SARIMA models for forecasting ZHVI trends
• Supports forecasting at different geographical levels (states, cities)
• Designed for low compute costs to maximize scalability
• Data processing and model training pipeline
• Comprehensive testing and evaluation of models
• Structured project for easy navigation and understanding
• Usage of Dataset with monthly updates(https://www.kaggle.com/datasets/robikscube/zillow-home-value-index)

Project Structure

1. models: Pre-trained SARIMA models for each geographic level
2. state-documentation: Documentation and guides for each state model
3. state-outputs: Forecast outputs and visualizations for each state
4. state-scripts: Data processing and model training scripts for each state
5. output-scripts: Ready to use, double click them and input how many months forward
6. PROCESS.md: Detailed explanation of the data processing and model training pipeline
7. MODEL.md: In-depth description of the SARIMA model architecture and hyperparameters
8. README.md: Overview and guide to using the forecasting tool
9. ZHVI.csv: Zillow Home Value Index dataset used for training and forecasting

Documentation

• README.md: This file, providing an overview of the project and its usage
• state-documentation: Detailed documentation for each state model
• PROCESS.md: Detailed explanation of the data processing and model training pipeline
• MODEL.md: In-depth description of the SARIMA model architecture and hyperparameters

Model Performance

The project includes SARIMA models for several states, each demonstrating high accuracy in forecasting ZHVI trends. The accuracy rates for the state models, sorted from highest to lowest accuracy, are as follows:

1. New York: 99.43%
2. New Jersey: 99.02%
3. Pennsylvania: 98.99%
4. Massachusetts: 98.11%
5. Illinois: 97.41%
6. Washington: 96.36%
7. North Carolina: 95.55%
8. California: 91.84%
9. Florida: 87.48%
10. Texas: 83.41%

These accuracy rates are derived from the model's performance on the test set, which is a portion of the data held out during training to evaluate the model's ability to generalize to unseen data.

Usage Guide

1. Clone the repository:

   git clone https://github.com/DimitriVavoulis/customizable-real-estate-market-forecasting-tool.git
   cd customizable-real-estate-market-forecasting-tool
   

2. Install dependencies:

   pip install pandas numpy matplotlib statsmodels joblib
   

3. Run the desired state's forecasting script. There are 10 state scripts available:

   • california-zhvi-sarima-forecast-script.py
   • florida-zhvi-sarima-forecast-script.py
   • illinois-zhvi-sarima-forecast-script.py
   • massachusetts-zhvi-sarima-forecast-script.py
   • new-jersey-zhvi-sarima-forecast-script.py
   • new-york-zhvi-sarima-forecast-script.py
   • north-carolina-zhvi-sarima-forecast-script.py
   • pennsylvania-zhvi-sarima-forecast-script.py
   • texas-zhvi-sarima-forecast-script.py
   • washington-zhvi-sarima-forecast-script.py

   For example, to run the California forecasting script:

   python california-zhvi-sarima-forecast-script.py
   

4. Follow the prompts to input the desired forecast horizon and view the output in the state-outputs directory.

For detailed usage instructions and customization options, refer to the state-specific documentation in the state-documentation directory.

Future Enhancements

• Expand coverage to more states and cities
• Implement a user-friendly web interface for easier interaction
• Integrate additional data sources for more comprehensive forecasts
• Explore other forecasting models and techniques for comparison

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.

Contact

For any questions, suggestions, or collaborations, please open an issue on the GitHub repository or contact Dimitri Vavoulis.