Also known as model decay or prediction drift, the characteristic that over time a given model that performs well against real-world data tends to perform worse over time as the real-world data and/or operational environment continues to change against the data under which the model was originally trained. Models can also drift due to changes in expectations for model performance, ways in which models are being applied, or other situations that may not be dependent on real-world data. The specific drift is the drift in predictions from expected and acceptable levels of performance to unexpected and/or unacceptable levels.
Login to register for events. Don’t have an account? Just register for an event and an account will be created for you!