Model accuracy is falling because incoming data no longer looks like the training data. What is this, and what should be automated?
Why B is the answer
Data drift, with monitoring that triggers an alert or retraining
Changed input distributions are data drift. Monitor for it and trigger retraining when thresholds are crossed.
In detail
- Inputs changing from the training data is data drift.
- Monitoring can trigger an alert or retraining.
- Check the retrained model before it replaces the old one.
Use case
An illustrative scenario, not a real incident.
A demand model weakens as customer behavior shifts. Drift monitoring raises an alert, and a retraining pipeline starts.
See it in different ways
flowchart TD A["Model monitoring"] --> B["Data drift: inputs changed"] A --> C["Prediction drift: outputs changed"] A --> D["Data quality: bad values"] A --> E["Performance: accuracy fell"]
stateDiagram-v2 [*] --> Trained Trained --> Registered: version and lineage Registered --> Deployed Deployed --> Monitored Monitored --> Retrained: drift alert Retrained --> Registered Deployed --> Archived: replaced
flowchart LR S0["Inputs change"] --> S1["Monitor detects drift"] --> S2["Alert or retrain"] --> S3["Evaluate new model"]
Key point: Monitor drift, and automate the response.
Why the other options are wrong
- A. Quota exhaustion, fixed by requesting a limit increase on the workspace
- Quota causes throttling, not falling accuracy.
- C. Overfitting, fixed by lowering the learning rate and retraining once
- Overfitting shows up during training.
- D. A networking fault, fixed by adding a firewall rule for the data source
- A network fault would give errors, not gradual decline.