« Back to Glossary Index

Drift detection is a method used to identify and monitor phenomena known as Data Drift and Concept Drift. Two approaches can be used for drift detection:

  • Offline approach: Historical data sets are compared with a more recent data set using hypothesis tests, which can confirm the variation of statistical properties between the two sets with a certain probability of error.
  • Online approach: Given a continuous data stream, algorithms capable of analysing statistical properties are employed. This is done by managing temporal windows, which can have either a predetermined or dynamically selected size.