Settlement Buildings

Research by Dr. Nikhil Kaza

According to UN-Habitat, about a third of the world’s population lives in irregular/informal settlements/favelas/shanty towns/slums. Whatever names these neighborhoods have around the world, these settlements are at the core of many urban regions; they provide affordable housing for urban poor but at high social and health costs. However, to study these phenomena, we need to know where these settlements are and how they change over time.

Nikhil Kaza is taking on the challenge of identifying and characterizing these irregular settlements in Bengaluru, India, a rapidly urbanizing city. This city has become a technology hub that spurs large economic development and population growth (~40% between 1991-2011) at great environmental and human cost. As of 2017, 10 million people are estimated to live in this city, and the Karnataka Slum Development Board estimates that there are 600 slums in the city. However, other non-governmental sources contend that this is a gross underestimate and puts this number much higher (~1500). According to unofficial estimates, a quarter to a third of the population of Bengaluru lives in these slums.

Dr. Kaza is working to identify these settlements using high-resolution satellite imagery. Using and blending multiple techniques, such as human image interpretation, machine learning ensembles, crowdsourcing validation and field verification, he is able to identify over 2500 non-trivial candidate neighborhoods in different categories, putting the number much closer to estimates of the non-governmental sources. Using high-resolution images (2m – 0.5m) from Digital Globe Foundation, and blending volunteered information from OpenStreetMap, as well as official and publicly available sources, his team is able to construct a picture of the built environment of the city of Bengaluru. One of the key challenges is to manage and process 8 billion data points to arrive at these conclusions.

Some features used to detect and monitor irregular settlements


Computers, while great at processing large data, are not yet able to quickly categorize complex information.T o weed out false positives identified by the machine learning algorithms, the team created a website ( where citizen scientists help validate if the images presented are irregular or not. By generating a large number of responses from a lot of users with different levels of experience, expertise, and accuracy, Dr. Kaza and his team hopes to both improve the accuracy of the machine learning algorithms and provide more current datasets for government and non-governmental agencies to target their policies and programs.


Soliciting crowdsourced verification to improve the machine learning algorithms


Not all irregular settlements are informal; the extent to which these settlements are outside the formal sector can only be determined through field validation and surveys and examination of public records. Currently field teams are deployed in these neighborhoods to conduct a detailed household level survey to understand how the settlements work and neighborhoods operate.

The plan is to scale up this work to cover all of India and perhaps all of the world. With the explosion of available high-resolution geospatial data, it is becoming possible to study these areas in greater detail.

Comments are closed.