A change in the statistical properties of input data over time, which can degrade a deployed machine learning model's performance.
A model trained on last year's customer behavior may quietly become less accurate as real-world patterns shift — monitoring for data drift catches this before it becomes a visible problem.
