Permutation Feature Importance is a technique used to assess the importance of a feature after a model is trained. It is defined as the reduction in model performance when the values of that feature are shuffled. The higher the performance drop after shuffling, the more important the feature is.