Ablation study

a formalization of variable isolation in machine learning that systematically removes or disables individual components of a model and measures how performance changes to determine each component's contribution