Using maps to predict economic activity

TL;DR

This paper presents a machine learning method that uses historical and contemporary maps to predict various economic and demographic statistics across different regions and time periods.

Contribution

The paper introduces a simple feature extraction algorithm from maps for predicting economic indicators, demonstrating its effectiveness on historical and modern datasets.

Findings

Maps can reliably predict population density in 1950s Sub-Saharan Africa.

Contemporary South Korean maps can predict income, consumption, employment, and electric consumption.

The method can also predict historical population growth in South Korea.

Abstract

We introduce a novel machine learning approach to leverage historical and contemporary maps and systematically predict economic statistics. Our simple algorithm extracts meaningful features from the maps based on their color compositions for predictions. We apply our method to grid-level population levels in Sub-Saharan Africa in the 1950s and South Korea in 1930, 1970, and 2015. Our results show that maps can reliably predict population density in the mid-20th century Sub-Saharan Africa using 9,886 map grids (5km by 5 km). Similarly, contemporary South Korean maps can generate robust predictions on income, consumption, employment, population density, and electric consumption. In addition, our method is capable of predicting historical South Korean population growth over a century.