Machine learning refers to a set of computer science techniques that allow computers to discover patterns in the data without being explicitly programmed. The U.S. Census Bureau has a rich history of using computational tools to learn about populations and the economy. Machine learning encompasses these methods, and also includes an additional set of highly efficient and effective modeling techniques that can be used to impute, classify, or predict patterns in data. For example, machine learning is routinely used by businesses for a wide variety of activities, including fraud detection, search relevance ranking, spam filtering, and self-driving cars.  Machine learning algorithms are also used, for example, to identify patterns in large amounts of data scraped from the web.  In doing so, large amounts of data can be analyzed more efficiently and effectively.

As the U.S. Census Bureau pushes into the 21st Century, the wealth of accessible data that can further its mission continues to grow. In order to make sense of the complex and voluminous data that we receive from various sources, we are using machine learning techniques to extract accurate insights from data in the most cost effective ways possible. Computers can discover hidden patterns among data more efficiently than humans, especially in the feature-rich data found in many big data sets. When big data sources are properly coupled with administrative and survey data, machine learning can serve to "impute" survey responses, help to reduce respondent burden and decrease costs. They can also help to build current and new Census Bureau products in a timelier manner.

Examples:

1. Estimating survey response propensity in Census blocks and tracts.
2. Classification of business establishments into specific NAICS codes.
3. Estimating population from satellite imagery.