In deployed machine learning pipelines, “drift” is changes in the model environment that cause the model performance to degrade over time. Drift might result from data quality changes. For example, increasing amounts of missing values in the input data. Or a company might alter the definitions in categories (e.g. product groupings) that are features in a model, or the relative proportions of records in those categories may change. “Concept drift” occurs when the relationships between predictors and the outcome changes. Drift is addressed through periodic re-training of the model.
