Why is a dataset divided into training and test portions?
Show answer and explanation
Answer: B
A model can reproduce patterns it has already seen without having learned anything general. Holding back a test portion gives an honest estimate of how the model behaves on unseen records, which is the situation it will face in production. The training portion is used to fit parameters, and the test portion is touched only for final evaluation. Leakage between the two, such as duplicated rows or preprocessing fitted on all data, inflates the result and defeats the purpose.
Why the other options are wrong
- A. Splitting does not reduce the work materially; it deliberately withholds data from training.
- C. Training on less data usually does not speed up the process in a meaningful way, and speed is not the objective.
- D. The split does not alter the model's structure; architecture is a separate design decision.
Exam tip. Holdout data gives an honest generalisation estimate.