Machine Learning for Land Use Scenarios and Urban Design

Geographic Information Systems (GIS) are becoming a more common tool in the practice of urbanism and urban design. Usually, GIS is used to visualize geo-located data to gain inside into the urban fabric, to either plan interventions within it, restructure it, or extend it. One problem for a data-driven planning process with GIS is how to turn the gained data into knowledge to drive a project. This paper discusses the use of super- and unsupervised machine learning to develop land-use scenarios for a vacant site within the city parameters of Berlin. Unsupervised learning is used to find cluster which shares certain characteristics. This interpretation of the data helps to make more informed decisions. As an example, for supervised learning, a neural network was trained to develop land-use scenarios fully autonomously. Autonomously generated land-use scenarios are an essential step to bridge the gap between the analysis and the design phase of urban development and enable the use of artificial intelligence in the planning process.