Perturbation methods create modified versions of a dataset by making small perturbations to data values. This might involve permuting values or adding small offsets to values. These variants can be used in emprical statistics, during training of machine learning algorithms, or as part of several techniques for explainable AI.
Also used in hcistats2e: Chap. 10: pages 115, 118
Also known as: perturbation-based methods, perturbations, perturbing individual values
Used in glossary entries: machine learning, permutation statistics
Figure 10.5: Different ways to modify a dataset.