Poverty maps are essential tools for governments and NGOs to track socioeconomic changes and adequately allocate infrastructure and services in places in need.
This interactive map tool is for exploring and comparing the inferred wealth in Sierra Leone and Uganda from multiple machine learning models. These models leverage seven independent and freely available feature sources based on satellite images, and metadata features collected via online crowd-sourcing and social media.
In this research, wealth is scored from 0 to 100. Households with score=0 have no assets and possess the lowest quality housing. In contrast, households with score=100 represent the richest end of the spectrum. On the maps, each location contains multiple households and is colored by the mean wealth. Most features were extracted in 2022. In the prediction view, models were trained with ground-truth data from the last two available surveys (2016 and 2018-19), and on the comparison view, the left panel was trained on 2016 and the right on 2018-19.
Feature-based: the CatBoost model, is trained with metadata features (i.e., Demographics, Mobility, Population, Nightlight intensity, Antennas, and Infrastructure).
Image-based: the Convolutional Neural Network model with daylight satellite images.
Combined: uses all features.
Weighted: “weighted samples” applied during training to correct for wealth imbalance.
Augmented: “offline augmentation” applied to correct for small samples.
Lisette Espín-Noboa, János Kertész, and Márton Karsai. 2023. Interpreting wealth distribution via poverty map inference using multimodal data. In Proceedings of the ACM Web Conference 2023 (WWW ’23), April 30-May 4, 2023, Austin, TX, USA. ACM, New York, NY, USA, 12 pages. https://doi.org/10.1145/3543507.3583862