Context-rich Urban Analysis Using Machine Learning: A Case Study in Pittsburgh, PA

This thesis reports on the analytic and generative potential of using machine learning methods on context-rich urban datasets. Working with a dataset of Pittsburgh's urban data, the paper proposes a novel method to extract information about each building in the city in ways that reflect its immediate urban context and analyze it in terms of urban homogeneity through machine learning algorithms. The method can identify urban characteristic or gradients that yield surprising correlations with Pittsburgh's distinctive neighborhoods by clustering feature vectors of the context-rich urban dataset. In the realm of urbanism, this thesis suggests the new definition of urban morphological types. This can be interpreted the new concept of urban fabrics based on the urban homogeneity with quantified investigation for qualitative traits of urban space and contexts, contributing to clarify the definition of urban fabrics which is a gray area of urban studies. This main contribution of this research indicates that artificial intelligence technology allows people to newly and unprecedentedly interpret the city space and triggers the architectural and urban design paradigms based on the interpretation.