Machine learning and phone data can improve targeting of humanitarian aid

Summary

Peer-reviewed Nature article evaluating the ML-based targeting approach used in Togo's Novissi programme. Reports that phone-based ML targeting reduced exclusion errors by 4-21% vs geographic targeting but increased errors by 9-35% vs hypothetical PMT. Provides detailed methodology for model evaluation including accuracy, precision, exclusion/inclusion error metrics, and fairness analysis across demographic subgroups. Documents use of gradient boosted trees and regularised logistic regression on CDR features.