I investigate how local governments segment their constituents through servicing their demands for public services. Using a dataset of 311 service requests from Miami-Dade County’s Open Data Hub, I looked for patterns among a number of important features measuring variation in constituents’ socioeconomic statuses, community-level civic engagement, types of services requested, length of projects, and geography. I used several unsupervised clustering algorithms (K-means, DBSCAN, Gaussian Mixture Models) to group the data and then explored the clusters produced by the best performing algorithm, using Silhouette Scores as a metric. The dataset had 319,855 service requests from county residents as of January 2020. The plot above shows a PCA two-dimensional plot of a K-means 2 cluster solution of the data (the best performing algorithm).
The project’s Jupyter Notebook is here, but in the meantime, below you may see a series of descriptive plots of features used in the analysis and plots exploring the clusters from the K-Means two cluster solution.