Grace Kim ’26, Computer Science

Grace Kim ’26

Major

Computer Science

Project Title

Machine Learning Applications to Identify Building Typologies in Minneapolis and New York City

Minor(s): Statistics and Machine Learning
Certificate(s): Optimization and Quantitative Decision Science

My project addressed the challenge of estimating large-scale embodied carbon emissions in the buildings and construction sector, which is responsible for over one-third of global carbon emissions. A building’s embodied carbon emissions, occurring during all lifecycle stages when the building is not in use, depend heavily on building type, but the lack of comprehensive data on these aspects has made it difficult to accurately estimation these embodied emissions. This project aims to contribute to more effective carbon reduction strategies by improving the accuracy of embodied carbon emission estimates. To tackle this, I developed scalable Deep Learning models using Convolutional Neural Networks to detect and validate building features critical for estimating building typology. Using Minneapolis Tax Parcel Data as the ground truth dataset, I trained the models on Google Street View images to predict these features and then assessed the models’ accuracy in predicting building features in Minneapolis. My models’ estimates were highly accurate: 88.7% for basement detection, 78.7% for differentiating between single-story and multi-story, 79.5% for detecting garage and 86.9% for predicting frame type. Finally, I evaluated the model’s scalability for application in New York City.