El sobreajuste ocurre cuando un modelo aprende los datos de entrenamiento demasiado de cerca, incluido el ruido o detalles idiosincráticos, y rinde mal con datos nuevos. Suele evidenciarse por un alto rendimiento en entrenamiento pero un rendimiento más débil en validación o prueba.
"The production of an analysis that corresponds too closely or exactly to a particular set of data, and may therefore fail to fit to additional data or predict future observations reliably". In other words, an overfitted model memorizes training data details but cannot generalize to new data. Conversely, an underfitted model is too simple to capture the complexity of the training data.
Creating a model that matches the training data so closely that the model fails to make correct predictions on new data. Regularization can reduce overfitting. Training on a large and diverse training set can also reduce overfitting. Overfitting is like strictly following advice from only your favorite teacher. You'll probably be successful in that teacher's class, but you might "overfit" to that teacher's ideas and be unsuccessful in other classes. Following advice from a mixture of teachers will enable you to adapt better to new situations. See Overfitting in Machine Learning Crash Course for more information.
In machine learning, creating a model that matches the training data so closely that the model fails to make correct predictions on new data.