Training features are the features (columns in a table, fields in records) that are chosen to be used in training a machine learning system. The raw training data may incude many features, some of which may need to be removed for privacy reasons, some because they are deemed irrelevent, and some in order to perform data reduction to improve or reduce the costs of training. The choice of training features can have radical effects on the eventual model and can be a source of bias.
Used in Chap. 9: page 116
Data reduction methods. Fig 11.1 from Statistics for HCI (2e).