Scientists at Stanford University have found a new method in predicting poverty through the use of machine learning and satellite images. The technique could make it easier for organizations to know where across the world their aid is needed most. Also, this could help governments develop a better policy to prevent or fight poverty.
Using three data sources namely daytime images, night light images, and survey data, scientists built an algorithm to predict how wealthy or poor an area is. The results of the study have been published in the journal Science.
"The idea is that if we train our models right, they help us predict poverty in areas where we don't have the surveys, which will help out aid orgs that are working on this issue," explained Neal Jean, co-author of the study and a doctoral candidate at Stanford.
The algorithm used in the study was built using a two-step process called "transfer learning," which is the most accurate way to predict the average consumption of a household and the wealth of villages. The first step involved deep learning techniques, where computers were taught to predict where night lights could be found simply by looking at daytime images. Through this method, researchers can better predict which areas are poor.
Next step involved a different model, which is the ridge regression model which knows the connection between lights and land features. Additional information was fed into the computer which was actual survey data from the World Bank Living Standards Measurement Study and Demographic Health Services.
"If you give a computer enough data it can figure out what to look for. We trained a computer model to find things in imagery that are predictive of poverty. It finds things like roads, like urban areas, like farmland, it finds waterways - those are things we recognize. It also finds things we don't recognize. It finds patterns in imagery that to you or I don't really look like anything... but it's something the computer has figured out is predictive of where poor people are," stated Dr. Marshall Burke.
NASA has been recording night lights since 2012. With the space agency's newest and more accurate satellites, further studies would offer better data. The next step in the study is to work on other countries to map poor areas across the world.