Predicting global poverty using satellite imagery and open-source data
Building a sustainable world by eliminating poverty
Almost a billion people worldwide currently live on less than US $2 dollars a day. In response to this alarming statistic, the United Nations set the eradication of poverty as its primary sustainable development goal, and have committed to achieve this by 2030.
In order for countries to successfully work towards this goal, they need to effectively track the efficacy and progress of poverty targeted initiatives. However, reliable economic and welfare data remains scarce and costly to obtain. This makes it challenging to effectively measure poverty related outcomes, and developing programs to address them.
In order to address this problem, we are developing a platform that will provide NGOs and policy makers with precise and granular estimates of welfare and other development outcomes, at a fraction of the cost of traditional survey based methods.
We do this by using powerful machine learning algorithms and huge amounts of open-source data to predict the distribution and spread of welfare, and other socioeconomic factors, with a high degree of accuracy.
Most machine learning approaches typically work best when lots of of labelled data is available. However, there are only a few sources where you can find reliable and comprehensive global data on welfare and socioeconomic conditions.
Inspired by the great work initiated by the team at Stanford's Sustainability and Artificial Intelligence Lab, we have developed an approach that combines cutting-edge machine learning and a comprehensive collection of open source data, to produce accurate and interpretable estimates for wealth measures for any location in the world. Our models learn the relationships between welfare and other useful features, which are available globally, thus allowing us to generate precise estimates, even in countries where survey data is unavailable.
High Accuracy Poverty Estimates
Based on an initial assessment across multiple African countries, results indicate that our model is strongly predictive of welfare and yield accuracy. There is an improvements of up to sixty percent over other current best methods. The following charts numerically illustrate the accuracy of our model for estimating welfare in Nigeria.
High Resolution MAPS
Using globally available and open-source satellite imagery, layering environmental, socioeconomic and conflict data, we are developing models that can generate highly accurate and granular estimates of welfare for any location in the world. As an example, here are our mapped welfare predictions for Nigeria.