Data Drift

Back to Model Monitoring

When the statistical distribution of input data changes over time compared to the training data. Causes model performance to degrade even if the model itself has not changed. Requires monitoring input feature distributions.


mlops monitoring data-drift