Data Drift

noun

say it: “DAY-tuh DRIFT” /ˈdeɪtə ˈdrɪft/

A change over time in the data a model receives, compared with the data it was trained on, which can make its predictions worse.

What is data drift?

Data drift is when the inputs a model sees in use change from the ones it learned on, such as a café's customers getting younger or prices rising across the board. The model's guesses can quietly get worse, so you compare new data with the training data and check the model's errors as answers come in.

import pandas as pd
trained_on = pd.Series([3.5, 4.0, 4.2, 3.8])    # cup prices in the training data
this_month = pd.Series([5.1, 5.4, 4.9, 5.2])
print(round(this_month.mean() - trained_on.mean(), 2))   # 1.28: prices have drifted up

In a sentence

  • “Data drift crept in when the café opened a second location near a college.”
  • “The monthly check flagged data drift in order sizes.”

Level: Journeyman · More AI Basics terms at this level

Updated