Data Driven Modelling and Decision Support for Slums

In the DYNASLUM project we  plan to develop a decision support system for slums in general, which will help guide experts when evaluating or designing policies to improve conditions within slums. This involves developing new computational methods for analyzing satellite images and new data visualization techniques for simulation steering. 

Through extending the existing decision support prototype, and applying it to a new domain of slum policy, the hope is to  generalize the current software. Therefore, this project  will aim to develop a reusable software framework for decision support and disseminate it to other users.

Pachucu Slum

Data analytics is required to extract features from the datasets available to the team. These include satellite imagery, survey data and the general Indian Census data. While there are official figures regarding the number of slums in India, these figures are often hard to verify due to the scale of the cities and the rate of growth. Researchers have already attempted to apply remote sensing techniques to slum detection with some success. Generally, there are two approaches to this, based either on objects (i.e., buildings) or on texture (i.e., pixels). The team plans to develop a new slum detection algorithm based on the fisher information matrix (FIM), connected-component labelling (i.e., blob-detection), and machine learning algorithms.

Source: http://bangalore.citizenmatters.in/

Source: http://bangalore.citizenmatters.in/

The DYNASLUM project will focus on developing an agent-based model of slum growth; the team have an initial model already and plan to extend this in the proposal. More specifically, we will construct a new discrete-choice model for the slum dwellers of Bangalore using micro data from 1970 –2011. Individual, household and neighbourhood characteristics, together with accessibility measures, will be used to model individual behaviour influencing the residential preferences of slum dwellers. The model will be capable of simulating a population of agents that make a decision to migrate to different slums in the city.

We will develop a new scenario generation algorithm to enable efficient exploration of the scenario space. Again, this part of the project connects to the SIM-CITY framework, and the expectation is that the scenario generation module will be of use in other domains and therefore will be translated back into the framework. While the envisaged visual analytics software is domain specific, we hope to build upon existing knowledge (such as eSight developed by NLeSC) and perhaps develop a set of re-usable libraries for visual analytics in the field of governance and public policy.