Attribution method that measures how much each input feature contributes to a model's prediction by integrating gradients along a path.