In machine learning, an epoch refers to one complete pass through the entire training dataset, which allows the model to learn from the data iteratively. Multiple epochs are usually required to optimize the model's performance, as each epoch updates the model parameters to minimize the error between the predicted and actual outputs.