Data Mining the City, Columbia University GSAPP | Fall, 2016

Professors:Danil Nagy

Collaborative WorkCarmelo Ignaccolo, Nickolaus J. Sundholm, Shuman Wu

The project aims to determine a mass model of a neighborhood without depending on the availability of GIS data but using Google street view images. The project takes machine learning method and train the network in a city where we have access to building data and Google street view images, then apply the model to a neighborhood without available information to estimate the average building height, street view brightness, perceived safety.

New York City has open sources including comprehensive GIS data and Google Street image. Therefore, this project uses a portion of Upper Westside in Manhattan as a study model to train the neural network, which can potentially predict building height of another neighborhood.

A 3D model of the neighborhood is created using actual building height from PLUTO data. The height information is visualized with each cross representing the roof level of a building, and each line showing the difference between adjacent buildings.

Using the generated height data, a predicted visualization model is created and presented in the same style, to compare with the actual model.

This visualization will be able to test the predicted data's accuracy while applying the training process to different neighborhoods. When a well developed neural network is formed, it can be used to predict building height in places where GIS data doesn't cover. This model can also contribute to research works when combined with other information such as street safety, land use, etc.