A review of machine learning and satellite imagery for poverty prediction: Implications for development research and applications

Abstract The field of artificial intelligence is seeing the increased application of satellite imagery to analyse poverty in its various manifestations. This nascent but rapidly growing intersection of scholarship holds the potential to help us better understand poverty by leveraging big data and recent advances in machine vision. In this study, we statistically analyse the literature in the expanding field of welfare and poverty predictions from the combination of machine learning and satellite imagery. Here, we apply an integrative review method to extract key data on factors related to the predictive power of welfare. We found that the most important factors correlated to the predictive power of welfare are the number of pre‐processing steps employed, the number of datasets used, the type of welfare indicator targeted and the choice of AI model. Studies that used stock measure indicators (assets) as targets achieved better performance—17 percentage points higher—in predicting welfare than those that targeted flow measures (income and consumption) ones. Additionally, we found that the combination of machine learning and deep learning significantly increases predictive power—by as much as 15 percentage points—compared to using either alone. Surprisingly, we found that the spatial resolution of the satellite imagery used is important but not critical to the performance as the relationship is positive but not statistically significant. These findings have important implications for future research in this domain and for anyone aspiring to use the methodology.