Machine learning maps help find poorest communities in Sri Lanka
Decision-Centered Evaluation of Machine Learning Poverty Maps Using Mobile Phone and Satellite Data
Computers and Society
Summary
Finding the poorest areas is important but expensive using surveys. The authors used machine learning with mobile phone and satellite data to predict poverty in Sri Lanka. They showed their models could find most of the poorest places better than using satellite data alone. The study also looked at how well these predictions work for isolated areas and how different ways of testing affect accuracy. This helps know if such maps can guide where aid should go, but doesn’t confirm if they exactly measure daily poverty.
What this means in practice
- •For aid targeting agencies: Identify poorest administrative units more effectively for allocating limited poverty relief resources in Sri Lanka using combined mobile and satellite data models.
- •For geospatial data teams: Improve evaluation of poverty predictions by testing geographic transfer and addressing errors in socioeconomically isolated communities.
Tested on one dataset.
Authors
Chanuka Algama, Merl Chandana, Viren Dias, Kasun Amarasinghe
Abstract
Identifying the poorest communities is essential for poverty alleviation, but household surveys and censuses are costly and infrequent. Machine learning offers alternative poverty estimates from mobile phone and satellite data, yet average prediction accuracy alone does not show whether maps support targeting under limited budgets. We apply a decision-centered evaluation to 13,985 Grama Niladhari divisions in Sri Lanka, combining call detail records (CDRs), remote sensing (RS), and CNN-derived Landsat 8 embeddings. We assess recovery of the poorest administrative units, compare random and spatially grouped validation, and examine errors in socioeconomically atypical communities. Against a census-derived asset index (PC1), Random Forest achieves Recall@25\% of 0.830, compared with 0.450 for nighttime lights. Random splitting raises recall by 4.1 percentage points relative to Divisional Secretariat Division (DSD)-grouped holdouts. The combined model recovers 86\% of the 25 poorest DSDs by PC1, versus 69\% for RS-only and 67\% for CDR-only models. Spatially isolated divisions have 10.8\% higher prediction error. Stronger isolation--error association in RS-only models is consistent with spatial smoothing, although its cause remains unconfirmed. These findings support evaluating poverty maps by targeting performance and geographic transfer, while recognising that agreement with an asset index does not establish consumption-poverty accuracy.