Drift takes two main forms. Data drift happens when the inputs change: a sensor ages, a camera is moved, a new product mix runs on a line. Concept drift happens when the relationship between inputs and the target changes, for instance after a machine is overhauled and its healthy vibration signature shifts. Either way, a model that performed well at launch starts missing events or raising false alarms.
Drift is the main reason MLOps practices exist for IoT fleets: monitor input distributions and prediction rates, compare against ground truth where available, and retrain on a schedule or on trigger. Anomaly detection models are especially exposed because what counts as normal changes with seasons and operating modes. Retraining then requires a reliable OTA update path to the devices.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.