Train-test split is a technique used to evaluate the performance of a machine learning algorithm. It divides the dataset into two subsets:
This technique is suitable when a sufficiently large dataset is available.
Source: Chapter 3, Section 3.2
Examiners expect two things: (1) a clear definition stating it is an evaluation technique, and (2) the role of each subset (train = learning, test = comparing predictions vs actual). Mentioning that it avoids overfitting is a bonus but not mandatory for 2 marks. Keep it concise.