Monitoring for Machine Learnings
ML monitoring is watching a deployed model after release: checking latency and errors, comparing live input data with training data to detect drift, tracking prediction quality when labels arrive, and logging experiments. This unit in the Production ML stage explains why models degrade even when code does not change.
Stage: Production ML · Part of the Machine Learning roadmap · Last reviewed 19 November 2026
What to learn in Monitoring
- Distinguish data drift from concept drift with an example
- Compute a drift measure such as population stability index on a feature
- Track live performance when delayed labels become available
- Log parameters, metrics and artefacts for each run using MLflow
- Set alert thresholds that trigger a retraining review
Monitoring interview questions
What is the difference between data drift and concept drift?
Data drift is a change in the distribution of input features, for example customers becoming younger. Concept drift is a change in the relationship between inputs and the target, for example fraud tactics evolving. Both can lower accuracy and need different responses.
Why track experiments in ML?
Models depend on code, data, parameters and random seeds. Experiment tracking records them with the resulting metrics so you can compare runs, reproduce a result and explain why one model was chosen. Tools such as MLflow store this history.
Where Monitoring fits in the Machine Learning roadmap
Monitoring belongs to the Production ML stage (intermediate level), which takes about 3 hours. From notebook to service.
Other units in this stage
Back to the full Machine Learning roadmap