Test set

How it is used

Data is often split into training, validation and test sets, for example 70/15/15. You hold out a test set and evaluate the model on it. If information from the test set leaks into training, results look better than they really are.

Example sentences

1. On the test set, the model's accuracy fell from 92% to 84%.

2. Keep the last three months as the test set so we can see how the model handles recent data.

Common collocations

• hold out a test set
• evaluate on the test set
• train-test split
• test set performance

Related terms

• training dataThe examples, with known inputs and usually known answers, that a model learns from; they are kept separate from the data used to test the model.
• cross-validationA way of testing a model by splitting the data into several parts, training on all but one, testing on the remaining part and repeating until each part has been used for testing.
• overfittingA problem in which a model learns the training data too closely, including its noise, so it performs well on that data but poorly on new data.
• accuracyThe share of all predictions that a model gets right; for example, 90 correct predictions out of 100 means an accuracy of 90%.

This word in other fields

The same English word can have a different meaning in another field:

• test set · Software Engineering
• test set · Artificial Intelligence & Machine Learning

More terms: Modelling and Machine Learning

• model
• feature
• regression
• classification
• precision
• recall
• baseline
• feature engineering
• target variable
• clustering