The Million Neighborhoods Map is a new tool designed to detect and identify informal settlements across the world. The tool, made by the University of Chicago’s Mansueto Institute for Urban Innovation, finds communities with limited access to roads and therefore other services. This means that the map can be used to identify urban areas most in need of roads, power, water, sanitation and other infrastructure.
On the Million Neighborhoods Map urban areas with limited access to road networks are colored red, while settlements with high access to streets are colored blue. If you zoom in on a city on the map you can therefore quickly identify those areas which are most likely to be informal settlements, areas which have sprung up with little planning and possibly little essential infrastructure.
The Million Neighborhoods Map doesn’t include much information about how the map was made. However reading between the lines of the ‘Interactive explainer’ on the map I suspect that machine learning has been used to identify informal settlements – particularly by identifying areas with a high density of buildings which have no direct access to roads or streets. This seems to be supported by the approach suggested in the paper The Fabric of Our Lives, one of whose authors is Luis Bettencourt, the inaugural Pritzker Director of the Mansueto Institute for Urban Innovation. This paper argues that
“We can take any city block and diagnose its degree of inaccessibility to each building from the transportation network using measures from graph theory and topology. On a larger scale, we can scan an entire city to identify blocks within which some buildings lack access.”
This appears to be the approach that has been taken to identify informal settlements in the Million Neighborhood Map.
Another example of using machine learning to identify informal settlements has been developed by Dymaxion Labs. Dymaxion Labs’ Maps of Potential Slums and Informal Settlements used machine learning to search the satellite imagery of a number of South American cities in order to identify and find slums and informal settlements. The resulting maps are being used to help urban planners and local councils identify where vital utilities need to be directed.
To help identify informal settlements Dymaxion Labs used the Random Forest machine learning technique. They applied the Random Forest technique to known informal settlements on satellite imagery from South American cities. The Random Forest classifier finds common features found in areas with known informal settlements and absent from areas without informal settlements. It then uses the classifier on new satellite imagery to automatically detect informal settlements in this satellite imagery.
The Machine Learning techniques developed for the Million Neighborhoods Map and by Dymaxion Labs can both be used by governments, local authorities and by non-profit agencies to identify informal settlements. These informal settlements spring up organically in urban areas with little central planning. They are therefore areas which often lack basic services, such as water, sanitation and electricity. Once the location of informal settlements has been identified vital infrastructure can be targeted at these areas. The Million Neighborhoods Map suggests that the same machine learning techniques that have been used to identify the informal settlements can also be used to work out the best way to improve access to streets and roads for the people living in these neighborhoods.