Using Machine Learning to Estimate Multidimensional Poverty

The Global MPI typically relies on Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS), however, these sources have limitations due to data gaps. In this seminar, speakers from the Human Development Report Office at UNDP present their paper which uses both daytime and nighttime satellite imagery data to develop machine-learning models that allow us to train the data at a local level (e.g., cluster level) to predict multidimensional poverty. The paper's analysis shows that satellite imagery at the cluster level has strong predictive power, explaining significant variations in poverty within countries. The model shows stronger predictive capabilities in rural areas compared to urban areas. Additionally, standard of living indicators such as electricity, housing, and cooking fuel are predicted with high accuracy. Overall, these findings highlight the potential of satellite imagery as a valuable tool for predicting poverty and informing policies in regions with limited ground-level survey data.

Registration details

• Join in person: no booking required, refreshments provided
• Join online: booking required via bit.ly/40wN491