Learn AI · Level 3 · Tech and tools · step 5 of 16 · word 2 of 2
What is model drift?
Model drift is when a model gets worse over time because the world it works in has changed since the data it learned from.
In plain English
A model learns from a snapshot of the past. When people’s habits, prices or the rules change, that snapshot goes out of date and the model’s guesses slowly get worse. Nothing in the code broke, which is why drift is easy to miss.
An example
In spring 2020, the pandemic changed shopping almost overnight. MIT Technology Review reported that models for stock, fraud and marketing, trained on normal behaviour, started making bad calls, and people had to step in.
Watch out
Drift is caught by watching a live model’s inputs and results, then retraining it on fresh data. That is a core job of MLOps. More in how a machine learning model is built.
Learn AI path · Level 3 · Tech and tools · step 5 of 16
Done? Next stepHow to run AI on your own computer →See all of level 3