In this report for the Office of the Under Secretary for Economic Affairs at the DOC, we implemented and compared various unsupervised learning methods for clustering U.S. metropolitan areas. Specifically, we examined the k-means clustering, hierarchical clustering, and spectral clustering algorithms.

We applied each of these clustering methods to three buckets of metropolitan level features: industry and occupational mix, local economic conditions, and social and demographic characteristics. Then, we created a dataset with the most similar pairs of metropolitan areas. This project provides a foundation for future research relevant to designing any place-based policies.