What is model drift and how do you handle it in Keras deployments?

Model drift is degradation in performance due to changing data distributions.

It occurs when real-world data deviates from training distribution. Handling includes retraining pipelines, monitoring metrics, and continuous evaluation using validation streams.