A0553
Title: An exact game-theoretic variable importance index for generalized additive models
Authors: Amir Khorrami Chokami - University of Cagliari (Italy) [presenting]
Giovanni Rabitti - Heriot-Watt University (United Kingdom)
Abstract: Generalized additive models (GAMs) are a widely adopted tool in statistical modeling. The problem of assessing variable importance in GAMs is addressed by introducing a variance allocation approach based on the Shapley value. A closed-form expression is derived for this importance index, allowing for efficient computation in high-dimensional settings and under general dependence structures. The practical implication is discussed that when a variable's importance is negligible, it can be safely eliminated from the GAM, simplifying the model. The case studies show that the Shapley values offer more informative insights than p-values in terms of ranking the importance of variables.
