Advances in Business Statistics, Methods and Data Collection

25Using Machine Learning to Classify Products for the Commodity Flow Survey

Christian Moscardi¹ and Benjamin Schultz²

¹U.S. Census Bureau, Economic Reimbursable Surveys Division, Business Development Staff, Washington, DC, USA

²U.S. Census Bureau, Economic Management Division, Washington, DC, USA

25.1 Background

25.1.1 Commodity Flow Survey (CFS) Background

The Commodity Flow Survey (CFS) is a joint effort between the U.S. Census Bureau and the Bureau of Transportation Statistics (BTS) within the U.S. Department of Transportation (USDOT). Policymakers at the national and local levels use these estimates to make infrastructure planning decisions. The survey's most widely used estimates are the domestic movement of goods by origin geography, destination geography, the product shipped, and mode of transportation. More information can be found on the Census Bureau and BTS websites (Bureau of Transportation Statistics 2021; U.S. Census Bureau 2020).

To collect data, every five years (most recently in 2017), the CFS surveys business establishments across the country and asks those businesses for a sampling of their shipment records throughout the year. This work will focus on two fields about product information that survey respondents were asked to provide for each shipment record they reported to the Census Bureau. The first was a free‐text description of the product being shipped – an example response might be “steel beams.” The second field asked respondents to provide the best‐fitting ...