You've learned how to calculate various metrics for classification and regression models. But applying these metrics correctly requires a specific approach to handling your data. Evaluating a model on the same data it was trained on gives an unreliable, often overly optimistic, estimate of its real performance on new, unseen data.

This chapter introduces the fundamental practice of data splitting. We will cover:

• Why evaluating on unseen data is essential for assessing generalization.
• The concept and purpose of training sets and test sets.
• The standard procedure for performing a train-test split.
• Common ratios and the role of randomness in splitting.
• The limitations of a single split and a brief introduction to the concept of cross-validation.

By the end of this chapter, you'll understand how to prepare your data properly to obtain a more trustworthy evaluation of your machine learning models.