The change in model input data over time, which can lead to model performance degradation if not monitored and addressed.
Occurs when the statistical properties of incoming data shift away from the training distribution - because of seasonality, user behavior changes, or external factors - causing models to make increasingly inaccurate predictions. Effective drift management includes automated detection (e.g., comparing feature distributions), defined thresholds for alerts, and retraining or recalibration workflows to restore performance.
An e-commerce retailer’s demand-forecasting model suddenly underpredicts sales of summer clothing due to an unexpected weather anomaly. A drift detector flags the shift in temperature-feature distribution, triggers retraining on recent data, and restores forecast accuracy before stockouts occur.
