A Machine Learning Approach for Counting Language Minority Groups in the United States

To help enforce the Voting Rights Act, the U.S. Census Bureau must estimate how many voting-age citizens struggle with English in jurisdictions across the country. Traditional statistical models used in 2021 had limitations—especially in small areas—so researchers turned to machine learning for better accuracy. This study introduces a new random forest model, tailored with a beta-binomial approach, that significantly improved estimates for language minority groups like Bangladeshi and Sri Lankan communities. The results show how advanced modeling can sharpen insights into community needs, even in areas with limited data.