The phenomenon where a machine learning model's predictive accuracy declines over time because the real-world input data no longer matches the statistical patterns in its original training data.

A foundational MLOps term used to describe software predictions becoming outdated. It drapes changing consumer habits or market shifts in high-tech navigational ocean drift language.

• Ocean Navigation Metaphors: Comparing changing customer behavior data to a ship drifting off-course in ocean currents.
• Data Maintenance Mask: Blaming "model drift" for failing to update database records with current customer info.
• House-Cleaning Reality: Like a calendar becoming outdated in January and calling it "temporal calendar drift."

Apollo 13 (1990s) scene where flight controllers monitor spacecraft navigation meters, noticing subtle trajectory drift away from the lunar return path.

The decline in software accuracy that occurs over time as real-world data and user habits change.

“Our platform monitors user inputs to detect model drift and trigger automated retraining.”

“We update our sales prediction model monthly to prevent drops in accuracy caused by changing customer buying habits.”