What is Model Monitoring

Definition

Model monitoring is the continuous tracking of a deployed machine-learning model's performance, inputs, and outputs in production to ensure it keeps behaving as expected, detecting issues such as accuracy decline, drift, or anomalies so models can be corrected before they cause harm.

« Back to Glossary Index

Detects performance decline and drift after deployment
Tracks inputs and outputs to catch anomalies in real time
Triggers alerts and retraining before bad predictions accumulate
Maintains trust by proving models still perform as intended

Real World Example

A lender monitors its credit model in production and is alerted when approval patterns shift and accuracy drops as economic conditions change, prompting a retrain before flawed decisions pile up.

FAQs

What does model monitoring track?

Prediction accuracy, input and output distributions, latency, and signals of drift or anomalous behaviour in production.

Why is monitoring necessary after deployment?

Models degrade as real-world data shifts, so monitoring catches decline before it harms decisions.

What action follows a monitoring alert?

Teams investigate and typically retrain, recalibrate, or roll back the model to restore reliable performance.