AI Dictionary › AI Fundamentals

Train-test split

The train-test split is the fundamental practice of dividing a dataset into two separate parts before training a machine learning model: one portion, usually the majority, is used for training, and a smaller portion is set aside as the test set, to be used only at the end to measure how well the model performs on data it has never seen during learning.

Definition

How it works

The split is typically done in proportions such as 80-20 or 70-30 between training and test, chosen randomly to avoid systematic bias, unless the data is time-based, in which case chronological order must be respected so the model is not shown future information during training. The test set must remain completely isolated throughout model development: consulting it too many times to adjust choices, a common mistake called data leakage, produces optimistic and misleading performance estimates.

Applications

It is the basic safeguard against overfitting: a model that performs great on training data but poorly on the test set has likely memorized the specific details of training instead of learning generalizable patterns. A third set, the validation set, is often added to choose hyperparameters without compromising the fairness of the final test set, or the whole scheme is replaced with cross-validation when available data is limited.

History & etymology

The practice of separating estimation data from verification data belongs to classical statistics, but it became a standard machine learning procedure starting in the 1970s and 80s, as the research community began systematically recognizing the risk of a model appearing accurate only because it was evaluated on the same data it was built from. The name literally describes the operation: splitting the data between train and test.

Related terms

• Cross-validation
• Dataset
• Overfitting
• Evaluation metrics
• Hyperparameter tuning

More in AI Fundamentals

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.