Machine learning approaches to targeting emergency COVID-19 assistance in developing countries
Presented by Joshua Blumenstock with Emily Aiken, Lupe Bedoya, Suzanne Bellue, Aidan Covile, Dean Karlan, Mushfiq Mobarak and Chris Udry
As COVID-19 spreads in low and middle-income countries, economic disruptions have left hundreds of millions without work or income.
To offset the pandemic’s most devastating effects, national policymakers and humanitarian organizations are scrambling to provide emergency assistance to those who need it most. But determining "those who need it most" is a difficult proposition in many developing countries, where official government registries are often incomplete and out of date.
We present empirical results from three developing countries that suggest that the targeting of humanitarian aid can be improved by incorporating non-traditional data from mobile phone networks and satellites.
Results from Afghanistan show that machine learning methods, as applied to mobile phone metadata, can predict program eligibility as accurately as survey-based measures of wealth and consumption. In Togo, we show that social protections guided by mobile phone data would more effectively reach low-consumption households than standard approaches. In Nigeria, we find that high-resolution poverty maps, built by applying deep learning to satellite imagery and other geospatial data, can improve the accuracy of geographic targeting.
Collectively, these results highlight the potential for data science methods to play an important role in future approaches to humanitarian response.
SoDa Labs webinar series
The SoDa Labs webinar series provides a platform for researchers around the world to present work that uses novel and alternative data and/or tools from data science and beyond to answer social science questions.
Event Details
- Date:
- 21 July 2020 at 9:00 am – 10:00 am
- Venue:
- Online
- Categories:
- Economics; Econometrics and Business Statistics; General
Description
Presented by Joshua Blumenstock with Emily Aiken, Lupe Bedoya, Suzanne Bellue, Aidan Covile, Dean Karlan, Mushfiq Mobarak and Chris Udry
As COVID-19 spreads in low and middle-income countries, economic disruptions have left hundreds of millions without work or income.
To offset the pandemic’s most devastating effects, national policymakers and humanitarian organizations are scrambling to provide emergency assistance to those who need it most. But determining "those who need it most" is a difficult proposition in many developing countries, where official government registries are often incomplete and out of date.
We present empirical results from three developing countries that suggest that the targeting of humanitarian aid can be improved by incorporating non-traditional data from mobile phone networks and satellites.
Results from Afghanistan show that machine learning methods, as applied to mobile phone metadata, can predict program eligibility as accurately as survey-based measures of wealth and consumption. In Togo, we show that social protections guided by mobile phone data would more effectively reach low-consumption households than standard approaches. In Nigeria, we find that high-resolution poverty maps, built by applying deep learning to satellite imagery and other geospatial data, can improve the accuracy of geographic targeting.
Collectively, these results highlight the potential for data science methods to play an important role in future approaches to humanitarian response.
SoDa Labs webinar series
The SoDa Labs webinar series provides a platform for researchers around the world to present work that uses novel and alternative data and/or tools from data science and beyond to answer social science questions.
Event Contact
- Name
- SoDaLabs@monash.edu
- Phone
- Organisation