ML-based geographic sampling frames miss transitory populations in fragile regions Pubblico

Abstract

Post-conflict environments often lack reliable survey data, complicating aid distribution. We studied 18 Iraqi communities, comparing machine learning and traditional methods for creating sampling frames. Using satellite imagery and Microsoft’s GlobalMLBuildingFootprints, we validated and manually added points in QGIS. An on-site survey recorded building conditions and conducted interviews in inhabited residences. Across 210.20 km2, we identified 61,603 valid buildings, visiting 1,225. Of these, 1,061 were inhabited. Comparing automated and manually-located building structures, 27.54% of manually added points were buildings inhabited by internally displaced persons (IDPs), while this rate was 20.62% for machine-learning-found structures. This statistically significant difference suggests machine learning methods for structure detection and its use to create geographic sampling frames may overlook transient populations, who are often the focus of aid and social assistance programs.

Table of Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Data. . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 ML-Based Building Footprints . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Manually Placed Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.1 Footprints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.2 Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.3 Geocoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.4 Sampling Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.5 Survey Administration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Results. . . . . . . . . . . . . . . . . . . . . . . . . . 10

5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 14

About this Honors Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Emory College
Department	Economics and Mathematics
Degree	B.A.
Submission	Honors Thesis
Language	English
Research Field	Economics, General
Parola chiave	English Fragile Sampling GIS
Committee Chair / Thesis Advisor	O'Connell, Stephen, Emory University
Committee Members	Hansman, Christopher, Emory University Gupta, Neha, Emory University

Ultima modifica

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	ML-based geographic sampling frames miss transitory populations in fragile regions ()	2025-04-08 18:39:44 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions