Machine Learning Fundamentals › Training vs Testing Data

What are best practices for Training vs Testing Data

Advanced

Answer

Always split data before doing any preprocessing, use random or time based splits depending on the problem, and never look at the test set until the model is fully finished.
Real-world example A bank builds a loan approval model on data from the last five years and tests it on the most recent six months of applications to see how it would have performed in practice.

Common follow-ups: What is a validation set, What is cross validation, What is overfitting