All Times ET
Program is Subject to Change
Integrating Machine Learning in the ACES Survey Instrument (308036)*Valerie Mastalski, US Census Bureau
Keywords: Machine Learning
The Annual Capital Expenditures Survey (ACES) of the U.S. Census Bureau provides relevant and timely estimates of capital investment in new and used buildings and other structures, machinery, and equipment by private, non-farm businesses located in the U.S. These data are critical to evaluate productivity growth, the ability of U.S. business to compete with foreign business, changes in industrial capacity, and measures of overall economic growth. In an effort to minimize misclassification ACES collects information on “Other” capital expenditures item. Respondents are asked to provide a description of these expenditures. These expenditures are then reviewed and reclassified as spending for either structures or equipment or as not applicable. This paper describes our efforts to use machine learning (ML) to aid in the categorization at the time of data collection of the “Other” capital expenditures item. This will minimize the inclusion of expenditures in the ”Other” category for items that should be reported as structures or equipment and significantly reduce the number of write-ins requiring manual review and recoding by the ACES staff. ACES is the first economic program to incorporate ML into the reporting instrument. We describe the ML model, the integration of a ML web service API into the instrument, respondent reaction to the ML web service, overall accuracy, and impact on the survey — early results shows that respondents moved 75.4 percent of the 3,338 write-ins to either buildings or structures.