基于机器学习的贫困地图中的空间异质性：模型在哪里表现不佳？（英）

Yating Ru, Elizabeth Tennant, David S. Matteson, and Christopher B. Barrett ADB Economics Working Paper Series Spatial Heterogeneity in Machine Learning-Based PovertyMapping: Where Do Models Underperform? Yating Ru, Elizabeth Tennant, David S. Matteson,and Christopher B. Barrett Yating Ru (yru@adb.org) is a natural resources andagriculture economist at the Sectors Department 2,Asian Development Bank. Elizabeth Tennant(ejt58@cornell.edu) is a research associate and a visitinglecturer; David S. Matteson (matteson@cornell.edu) No. 798 | September 2025 TheADB Economics Working Paper Seriespresents research in progress to elicit commentsand encourage debate on development issuesin Asia and the Pacific. The views expressedare those of the authors and do not necessarily Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) © 2025 Asian Development Bank6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Some rights reserved. Published in 2025. ISSN 2313-6537 (print), 2313-6545 (PDF)Publication Stock No. WPS250340-2DOI: http://dx.doi.org/10.22617/WPS250340-2 The views expressed in this publication are those of the authors and do not necessarily reflect the views and policiesof the Asian Development Bank (ADB) or its Board of Governors or the governments they represent. ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for anyconsequence of their use. The mention of specific companies or products of manufacturers does not imply that they By making any designation of or reference to a particular territory or geographic area in this document, ADB does notintend to make any judgments as to the legal or other status of any territory or area. This publication is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO)https://creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be boundby the terms of this license. For attribution, translations, adaptations, and permissions, please read the provisions This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributedto another source, please contact the copyright owner or publisher of that source for permission to reproduce it. Please contact pubsmarketing@adb.org if you have questions or comments with respect to content, or if you wishto obtain copyright permission for your intended use that does not fall within these terms, or for permission to use Notes:In this publication, “$” refers to United States dollars. ABSTRACT Recent studies harnessing geospatial big data and machine learning have significantlyadvanced poverty mapping, enabling granular and timely welfare estimates in traditionally data-scarce regions. While much of the existing research has focused on overall out-of-samplepredictiveperformance,there is a lack of understanding regarding where such modelsunderperformand whether key spatial relationships might vary across places. This studyinvestigates spatial heterogeneity in machine learning-based poverty mapping, testing whether Keywords:poverty mapping, machine learning, spatial models, East AfricaJEL codes:C21, C55, I32 1.INTRODUCTION “To end poverty in all its forms everywhere by 2030” ranks first among the SustainableDevelopment Goals (United Nations General Assembly 2015). Nevertheless, years of progress inpoverty reduction have been disrupted by the coronavirus disease pandemic, violent conflicts,environmental shocks, sharp food price rises, and increasing global inequality both within andamong nations. After declining for three decades from 1990, in recent years both the number andthe percentage of people living in extreme poverty have risen globally (Lakner et al. 2022). In While traditional poverty estimation relying on surveys and census data are time consumingand resource-intensive, recent advances in geospatial big data, machine learning (ML) algorithms,and computational capabilities now enable poverty and wealth mapping at a higher resolution andin a timelier manner (Blumenstock, Cadamuro, and On 2015; Jean et al. 2016; Pokhriyal andJacques 2017; Yeh et al. 2020; Browne et al. 2021; Chi et al. 2022; Lee and Braithwaite 2022).The comprehensive global coverage of remote sensing-derived datasets has propelled these new Early poverty maps exploited Small Area Estimation (SAE), which combines sparse surveydata with dense auxiliary data (such as census or administrative records) to estimate poverty forsmall areas, linked by common variables (Ghosh and Rao 1994; Rao 1999; Elbers, Lanjouw, andLanjouw 2003; Christiaensen et al. 2012). Recently, SAE methods have integrated remotesensing and other geospatial data as additional sources of auxiliary data, using such data fusionto improve estimation accuracy, particularly for the poorest (Masaki et al. 2020). ML-based povertymapping is analogous to SAE but offers greater flexibility and spatial coverag