devscore 🌍

devscore is a Python package that computes a multi-dimensional Development Score using open geospatial, satellite, and survey-based indicators. It is inspired by cutting-edge development economics research from MIT, Oxford, and the World Bank.

It helps researchers, governments, and NGOs estimate local well-being and market access for any region in Africa.

Features

• Poverty Prediction: ML-based poverty estimation using satellite imagery, nightlights, and infrastructure data
• Market Access: Distance and travel time calculations to economic centers, markets, and key services
• Infrastructure Mapping: Density analysis of schools, hospitals, roads, and financial services from OSM
• Food Security Assessment: NDVI-based vegetation and agricultural productivity analysis
• Night-time Lights: Economic activity proxy using VIIRS satellite data
• Mobile Money Potential: Financial inclusion indicators

Installation

pip install devscore

Anaconda/Conda Installation (Recommended)

For better compatibility with geospatial dependencies, especially on Windows:

conda create -n devscore python=3.10 -y
conda activate devscore
conda install -c conda-forge geopandas rasterio osmnx scikit-learn xgboost -y
pip install h3 earthengine-api devscore

See CONDA_INSTALLATION.md for detailed conda setup instructions.

Install from Source

git clone https://github.com/idrissbado/devscore.git
cd devscore
pip install -e .

Dependencies

pip install numpy pandas geopandas rasterio earthengine-api osmnx scikit-learn xgboost requests h3 shapely

Quick Start

from devscore import compute_development_score

# Compute development score for a location
lat, lon = -1.2921, 36.8219  # Nairobi, Kenya
score = compute_development_score(lat, lon)

print(f"Development Score: {score['overall']:.3f}")
print(f"  - Poverty Score: {score['poverty']:.3f}")
print(f"  - Market Access: {score['market_access']:.3f}")
print(f"  - Infrastructure: {score['infrastructure']:.3f}")
print(f"  - Food Security: {score['food_security']:.3f}")
print(f"  - Mobile Money: {score['mobile_money']:.3f}")

Using Dynamic Weights

Instead of fixed weights, use data-driven or expert-based methods:

from devscore.scoring.final_score import DevelopmentScoreCalculator

# Method 1: AHP (Analytical Hierarchy Process) - Expert-based
calculator = DevelopmentScoreCalculator(weight_method='ahp')
result = calculator.compute_development_score(lat, lon)

# Method 2: Entropy - Data-driven, based on variation
calculator = DevelopmentScoreCalculator(weight_method='entropy')

# Method 3: Auto - Robust average of multiple methods
calculator = DevelopmentScoreCalculator(weight_method='auto')

# Method 4: Custom weights
custom_weights = {
    'poverty': 0.40,
    'market_access': 0.25,
    'infrastructure': 0.20,
    'food_security': 0.10,
    'mobile_money': 0.05
}
calculator = DevelopmentScoreCalculator(custom_weights=custom_weights)

See examples/dynamic_weights.py for comprehensive examples.

Data Sources

All data sources are open and freely accessible:

• Satellite Imagery: Sentinel-2 (AWS Open Data), Landsat 8/9
• Night-time Lights: NOAA VIIRS DNB
• Infrastructure: OpenStreetMap via OSMnx
• Population: WorldPop
• Wealth Data: DHS (Demographic and Health Surveys)

Methodology

The package implements a weighted aggregation of five key development indicators.

Weighting Methods:

The package supports multiple approaches to determine indicator weights:

1. Fixed Weights (Legacy): Traditional research-based weights
2. AHP (Analytical Hierarchy Process): Expert judgment-based (default)
3. Entropy Method: Data-driven, based on information content
4. PCA (Principal Component Analysis): Variance-based weighting
5. CRITIC: Combines variability and correlation structure
6. Auto: Robust average of multiple methods

Default AHP Formula:

Development Score = w₁ × Poverty + w₂ × Market Access + 
                   w₃ × Infrastructure + w₄ × Food Security + 
                   w₅ × Mobile Money

Where weights (w) are determined using AHP based on development economics research consensus. For fixed weights: w₁=0.35, w₂=0.20, w₃=0.20, w₄=0.15, w₅=0.10.

Each component is normalized to a 0-1 scale where 1 indicates highest development.

Components

1. Poverty Score: Machine learning model trained on DHS wealth index, using features like nightlights, NDVI, built-up area, population density, and road density

2. Market Access: Exponential decay function based on distance to markets, roads, hospitals, and economic centers

3. Infrastructure Index: Density of amenities (schools, clinics, markets, banks) per km²

4. Food Security: NDVI-based vegetation health indicator

5. Mobile Money: Financial inclusion proxy based on agent density and connectivity

Research Background

This package is inspired by academic research from:

• World Bank Development Economics (DEC)
• MIT Poverty Lab
• Oxford Centre for the Study of African Economies (CSAE)
• LSE International Growth Centre (IGC)
• Harvard Growth Lab
• Nature papers on satellite-based poverty prediction

📄 Read the full methodology paper: PAPER.md - A comprehensive 8000+ word article explaining the theoretical framework, algorithms, validation results, and practical applications.

📚 Additional Documentation:

• Conda Installation Guide
• Quick Start Guide
• Example Scripts

Citation

If you use this package in your research, please cite:

@software{devscore2025,
  title={DevScore: An Open-Source Framework for Multi-Dimensional Development Assessment Using Geospatial Data},
  author={Bado, Idriss Olivier},
  year={2025},
  url={https://github.com/idrissbado/devscore}
}

License

MIT License

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

Acknowledgments

Thanks to the open data community and organizations providing free access to geospatial data for development research.