Mapping the Dimensions of Poverty Through Big Data, Socioeconomic Surveys and Machine Learning in Cambodia

Cambodia has witnessed rapid economic growth in recent years; however, it remains one of the most economically vulnerable nations in Southeast Asia, grappling with persistent poverty challenges. Accurately understanding the multiple dimensions of poverty is essential for promoting sustainable development and guiding targeted policy interventions. Yet, traditional poverty data are often outdated and lack the granularity needed for effective subnational planning. To address this gap, this study leverages new big data sources, machine learning techniques, and the Cambodia Socio-Economic Survey (CSES) to predict and map multidimensional poverty across 10 indicators in three dimensions based on the Global Multidimensional Poverty Index (MPI): education, health, and living standard dimensions at fine spatial scales. By integrating deprivation probabilities across a gridded landscape with building footprint information, the study estimates household-level deprivations. Using a random forest algorithm, the study achieves high predictive accuracy for indicators such as clean water, sanitation, food consumption, housing materials, cooking fuel, and access to electricity. However, challenges remain, including the need for unbiased training data and the limited capacity to capture disparities within regional aggregates (provinces, districts, townships). Despite these limitations, the study identifies nighttime lights, population density, and road network data as key predictors of poverty. The findings demonstrate the feasibility of using big-earth observation data and machine learning to complement traditional socioeconomic surveys, enabling a more detailed and dynamic understanding of multidimensional poverty at various geographical scales.