In recent years, there has been significant progress in methods for creating poverty maps, much of which relies on the usage of advanced machine learning algorithms utilized with remote sensing data. Poverty maps created through this method are typically validated through comparing subnational poverty estimates gathered by machine with poverty assessments based on surveys. These survey-based estimates often provide a good evaluation of actual poverty level, but still could be impure. To investigate the accuracy of existing poverty map validation measures, a study used a pseudo-census created from the Mexican Intercensal Survey of 2015. Results of this study show that the validation process commonly used with machine learning approaches were misleading in terms of interpreting how models truly perform. Additionally, the results indicate that estimates based on machine learning techniques rival older than more data-intensive poverty mapping techniques. When assessing accuracy against the “true” poverty rate, the estimates from the publicly available geo-referenced data was found to do not well and underperformed older poverty mapping methods.
Person mentioned in this article: The person mentioned in this article is researcher Haydée Charur who conducted several design-based simulation experiments to investigate the accuracy of existing poverty map validation measures.
Company mentioned in the article: The company mentioned in this article is the Mexican Intercensal Survey of 2015, a non-profit organization used to help survey poverty around the globe. This organization helps to provide valuable data to governments, organizations, and research scholars looking to better understand poverty around the world.
