What is Data Drift?

Data Drift is the change in the data distribution over time on which a machine learning model operates, compared to the data it was trained on. This can lead to a decrease in prediction quality, as the model encounters new, unfamiliar patterns.

For example, if a model predicts product demand and new trends or seasonal changes appear in the data, which the model was not trained to account for, its accuracy may decline.

Statistical methods such as distribution comparison tests (e.g., KS test) and monitoring model metrics in production are used to detect Data Drift. When drift is detected, retraining the model on new data is often required.