The idea of using extreme value regression as an underlying model for backward selection in unbalanced setups is interesting. A combination with bootstrapping is highly relevant for real-world scenarios where datasets are often affected by low sample...