Neighbourhood detection with analytical tools

Abstract

The increasing population size of cities makes the urban fabric ever more complex and more disintegrated into smaller areas, called neighbourhoods. This project applies methods from geoscience and software engineering to the process of identification of those neighbourhoods. Neighbourhoods, by nature, are defined by connectivity, centrality and similarity. Transport and geospatial datasets are used to detect the characteristics of places. An unsupervised learning algorithm is then applied to sort places according to their characteristics and detect areas with similar make up: the neighbourhood. The attributes can be static like land use or space syntax attributes as well as dynamic like transportation patterns over the course of a day. An unsupervised learning algorithm called Self Organizing Map is applied to project this high dimensional space constituting of places and their attributes to a two dimensional space where proximity is similarity and patterns can be detected - the neighbourhoods. To summarize, the proposed approach yields interesting insights into the structure of the urban fabric generated by human movement, interactions and the built environment. The approach represents a quantitative approach to urban analysis. It reveals that the city is not a polychotomy of neighbourhoods but that neighbourhoods overlap and don't have a sharp edge.