Early stopping: Difference between revisions

[[Machine learning]] seeks to [[train]] a model that can make accurate predictions on new [[data]]. Unfortunately, during [[training]] it is common for the model to overfit the training data; that is, it becomes too complex and includes irrelevant details or noise in the dataset. Unfortunately, overfitting can lead to poor performance when faced with new data - thus defeating its purpose. Early stopping is an approach used in machine learning to prevent overfitting and improve generalization by monitoring model performance during training and stopping when its performance on validation sets starts to deteriorate.