|
Abstract:
|
This roundtable discussion will foster a discussion about training machine learning algorithms on survey data from complex sampling designs. Which machine learning algorithms or models are able to take complex survey design features into account? During parameter tuning, how should sample splitting or cross validation be adapted to account for non-iid designs? What pitfalls may occur when the survey design is ignored? What are some success stories, whether from government or industry or academia, of data products built on machine learning appropriately applied to survey data? In which areas is there a particular need for more research or better software?
|
