Overfitting

When a model learns the training data’s noise and quirks instead of the general pattern — it looks great on training metrics and performs worse on new, unseen data.