Alexandria, VA – August 20, 2020 – Gunnison along with our partner E3 Federal Solutions explores the use of Artificial Intelligence for the Census Bureau. Gunnison developed the U.S. Census Bureau’s software to support quality control for field operations during the 2020 Census for Address Canvassing (AdCan) and Non-Response Follow Up (NRFU) operations. Using an unsupervised clustering approach we explored how an Artificial Intelligence (AI) approach can make outlier detection more efficient and be used to identify new correlations.
We implemented our approach using Dataiku Data Science Studio with a PostgreSQL database to enable review and manipulation of data, and applied multiple machine learning functions on the data. We were able to identify new correlations that were not originally apparent such as long/short durations at odd times of day, duration correlation with renters vs. owners, and duration correlation with primary language.
Our approach did not require any specific tooling and was able to be readily implemented into the existing software baseline. Our open source tool selection provided low-overhead and quick turn-around. We produced similar results using AI in significantly less time and provided the ability to experiment with new tests quickly. Our approach can be implemented in a variety of forms – working with existing infrastructure if desired.
Clustering algorithm identified outliers and was used to quickly confirm correlation of longer durations for renters vs. owners. Other cluster variables allowed for visualization of duration over time, showing longer durations later in the day.
Figure: interview Duration for Renters vs. Owners by Interview Time of Day (Simulated Data)
Founded in 1994, Gunnison is celebrating 25 years as a certified small business. Gunnison takes pride in tackling its clients’ most ambitious IT projects and serving its critical mission objectives. Other major customers include the U.S. Patent and Trademark Office and the Census Bureau.