Uncovering commercial activity in informal cities

Area of Science:

• Urban studies
• Geographic Information Systems (GIS)
• Computer vision

Background:

• Understanding urban economic activity is crucial for policy, but data is scarce in developing cities with high informality.
• Street imagery and machine learning offer a novel, cost-effective solution for mapping economic landscapes.

Purpose of the Study:

• To develop and apply an algorithm for detecting visible establishments from street view imagery.
• To uncover and analyze previously unmapped economic activity in Medellín, Colombia.

Main Methods:

• Utilized machine learning algorithms to identify visible establishments in street view images.
• Applied spatial analysis to the generated dataset of economic activity.
• Compared the density of detected establishments with registered business data.

Main Results:

• Detected a polycentric urban structure with five distinct economic clusters in both central and peripheral areas.
• Inferred that informal economic activity concentrates in impoverished, densely populated neighborhoods.
• Highlighted a substantial discrepancy between official economic data and observed on-the-ground activity.

Conclusions:

• Machine learning applied to street imagery provides a powerful tool for understanding urban economies, especially in data-scarce environments.
• The study reveals the prevalence and spatial distribution of informal economic activities, crucial for informed urban planning.