The decay of a deployed model's performance over time, also called model decay, because the world it scores moves away from the data it learned from. Data drift and concept drift are its usual causes; a broken data pipeline can look the same.
Card 1 of 228. Answer: The decay of a deployed model's performance over time, also called model decay, because the world it scores moves away from the data it learned from. Data drift and concept drift are its usual causes; a broken data pipeline can look the same.
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
Model drift is the symptom: accuracy, AUC or error gets worse after deployment. The causes sit underneath it. Data drift means the inputs changed, concept drift means the relationship between inputs and target changed, and data quality problems mean the model is being fed something different from what the pipeline promises.
Example: a churn model launches with an AUC of 0.82 and a year later scores 0.71. Digging in finds three things. A price increase changed who leaves at a given usage level, which is concept drift. A new student plan brought in users unlike any in training, which is data drift. And a renamed plan code now arrives as missing for 8% of rows, which is a bug that no retraining will fix.
So the first step after a drop is separating the causes: check data quality and schema, then input drift, then performance by segment. Interviewers then ask how to catch it early (monitor performance, or estimate it while labels are delayed) and how to respond (fix the pipeline, retrain on recent data, or rethink features). Not all drift hurts the model, and not every drop in performance is drift.
