Abstract by Robert Alger
The United States Census and Machine Learning: A Practical Application
The United States Census Bureau released to the public its application programming interface (API) in 2012. The resulting population models and analyses have proven invaluable in supplementing research in nearly every branch of social science. However, much of the data remains confined to ink and paper, particularly data pre-1990. The result is an incomplete picture that has traditionally been filled by the arduous task of indexing -- requiring an individual to digitize handwritten information. Advances in machine learning provide an increasingly cost-effective and time-efficient solution.
Keywords: Machine learning, Neural netowrks, Handwriting recognition, Census data