In machine learning, what does “Overfitting” refer to?

Overfitting occurs when a machine learning model learns not only the underlying patterns in the training data but also the noise and outliers to a degree that adversely impacts its performance on unseen data. A model that is too complex relative to the dataset can easily capture noise, leading it to make predictions based on irrelevant patterns. This usually results in high accuracy on the training dataset but poor generalization to new, unseen data.

In contrast, options that suggest a model performing poorly on training data, or performing exceptionally on new data, do not align with the concept of overfitting. Overfitting specifically refers to the model's excessive complexity in relation to the dataset, leading it to memorize rather than learn.