An end-to-end real estate market analysis and forecasting engine for Riyadh and Jeddah. This project utilizes geospatial intelligence (Uber's H3) and is designed to evolve from statistical baselines into deep neural architectures.
To maintain engineering rigor, this project follows a phased development roadmap. The "Neural Engine" is the target state, built upon validated geospatial and statistical foundations.
- Phase 1 (Current): Data Acquisition & Baseline Modeling (Linear Regression, ~0.64 R²).
- Phase 2 (In-Progress): Geospatial Feature Engineering (H3 Multi-resolution indexing).
- Phase 3 (Upcoming): Deep Neural Network (DNN) implementation for non-linear valuation.
.
├── data/ # Dataset management (GitIgnored)
├── docs/ # Technical roadmap and documentation
├── notebooks/ # Exploratory Data Analysis & Cleaning
├── reports/
│ ├── figures/ # Model metrics and visualizations
│ └── interactive/ # Geospatial HTML maps
├── src/ # Core Production Logic
│ ├── data/ # Scrapers and preprocessors
│ ├── models/ # Baseline and planned DNN models
│ └── visualization/ # Map builders and graphing utilities
├── requirements.txt # Dependency list
└── README.md
- Hybrid Cleaning: Advanced outlier filtering using H3 geospatial resolution + room-based grouping.
- Interactive Visualization: Multi-layer maps with property-level markers and district-level heatmaps.
- Baseline Model: A validated Linear Regression regressor accounting for ~64% of price variance.
- Clone & Install:
pip install -r requirements.txt
- Data Setup: See data/README.md for instructions on where to place your local datasets.
- Execution:
- Run baseline training:
python src/models/baseline_regressor.py - Generate maps:
python src/visualization/map_builder.py
- Run baseline training:
This project is licensed under the MIT License.