Drift in Characterizations of Data
摘要
This chapter presents several examples from the authors’ research on applications of drift detection in industrial settings. These include representing a dataset as “slices” based on feature intervals, characterized by observation density or ML model performance, and modeling polynomial regression relationships between dataset features to detect statistical change in these relationships’ strength. The examples illustrate the usefulness of representing data in intermediate forms (e.g., slices, polynomial relationships) and detecting drift on these forms.