Machine learning and phone data can improve targeting of humanitarian aid

Area of Science:

• Data Science
• Development Economics
• Humanitarian Logistics

Background:

• The COVID-19 pandemic exacerbated food insecurity and poverty in low- and middle-income nations.
• Social assistance programs have reached over 1.5 billion people globally.
• Effective targeting of aid to those most in need is a persistent challenge.

Purpose of the Study:

• To investigate the efficacy of mobile phone network data in enhancing the targeting of humanitarian assistance.
• To develop and evaluate machine-learning algorithms for poverty identification using mobile data.

Main Methods:

• Utilized traditional survey data to train machine-learning algorithms.
• Applied trained algorithms to mobile phone data to identify patterns of poverty.
• Evaluated the approach using a COVID-19 emergency cash transfer program in Togo.

Main Results:

• The machine-learning approach reduced exclusion errors by 4-21% compared to geographic targeting.
• Compared to a hypothetical comprehensive social registry, exclusion errors increased by 9-35%.

Conclusions:

• Mobile phone data offers a valuable complement to traditional methods for humanitarian aid targeting.
• This approach is particularly relevant in crisis situations with incomplete or outdated data.