High-resolution socioeconomic data are crucial for place-based policy design and implementation, but it remains scarce for many developing cities and countries. We show that an easily accessible and timely updated neighborhood attribute, restaurant, when combined with machine-learning models, can be used to effectively predict a range of socioeconomic attributes. This approach allows us to collect training samples from representative neighborhoods and then use our trained model to infer unsampled neighborhoods in the city in a granular, timely, and low-cost manner. The good cross-city transferability performance of our model can also help bridge the “data gap” between cities, by training the model in cities with rich survey data and then applying it to cities where such data are unavailable.
Predicting neighborhoods’ socioeconomic attributes using restaurant data
Lei Dong, Carlo Ratti, and Siqi Zheng