Glossary

Training data

The dataset with known outcomes that a machine learning model learns patterns from before being used to predict new cases.

Also called: training set

Training data is the dataset a model learns from: a set of examples, each with input features and, for supervised learning, a known correct outcome, that the model uses to learn the pattern between them. The quality, volume and representativeness of training data generally matters more to a model's real-world performance than the choice of algorithm.

Before training, this data is typically split so that not all of it is used for learning: a train-test split holds out a portion to check how the trained model performs on data it has never seen. Training data also needs cleaning and shaping through feature engineering, and its labels, where a human or a process marks the correct outcome, must be accurate, since a model trained on mislabeled examples learns those mistakes as if they were real patterns.

Two recurring pitfalls dominate discussions of training data. First, it must reflect the population the model will actually be used on; data that is unrepresentative, for example skewed toward one region or time period, produces a model that performs poorly or unfairly elsewhere. Second, if information that would not be available at prediction time leaks into the training set, the model can appear highly accurate while learning a shortcut that will not exist in production, a subtle form of overfitting.

Last reviewed September 22, 2026

In the index now

Related terms

Train-test split Dividing a dataset into separate portions for training a model and for evaluating it, so performance is measured on unseen data. Feature engineering Creating, transforming or selecting the input variables a model is trained on, to make the underlying pattern easier for it to learn. Overfitting When a model learns the noise and specific quirks of its training data so closely that it performs poorly on new data. Supervised learning A machine learning approach where a model learns from examples that already include the correct answer, to predict that answer on new data. Data quality The degree to which data is accurate, complete, consistent and fit for the purpose it is being used for.

Related guides

How to choose a data labeling tool Choosing a labeling tool means choosing between a UI you staff yourself, a vendor's workforce, or skipping manual labeling altogether. Image and video analytics, explained How pictures and video become structured data — computer vision, the labeled training data it depends on, and how streaming teams measure watch quality.