developer-roadmap/src/data/roadmaps/mlops/content/107-mlops-components/106-monitoring.md

Monitoring and Observability

Monitoring in MLOps primarily involves tracking the performance of machine learning (ML) models in production to ensure that they continually deliver accurate and reliable results. Such monitoring is necessary because the real-world data that these models handle may change over time, a scenario known as data drift. These changes can adversely affect model performance. Monitoring helps to detect any anomalies in the model’s behaviour or performance and such alerts can trigger the retraining of models with new data. From a broader perspective, monitoring also involves tracking resources and workflows to detect and rectify any operational issues in the MLOps pipeline.

• ML Monitoring vs Observability article
• Course: Machine learning monitoring concepts, Monitoring ML in Python
• Tools: Prometheus, Grafana