Model drift refers to the decline in a model’s predictive accuracy over time due to changes in the real-world environment. This phenomenon occurs when the relationships between variables in the data evolve, rendering the model’s assumptions less effective. Causes include shifts in user behavior, technological updates, or alterations in data distribution.
For example, a spam detection model trained on specific email content may experience drift when spammers change their messaging strategies, making the model less effective at identifying spam.