It involves identifying extreme or aberrant data points that may not be representative of the dataset as a whole.