Game theoretical approach to determine feature importance. A practical case of use

D. Samaniego, L. Davila-Pena

In the last years, the usage of Shapley value is being extended through machine learning algorithms as a method to determine the importance of a feature on its contribution to a certain target value. This is done by defining a cooperative game where features play the role of voters.

In this talk we will comment an application of this method into e-commerce business decisions where the target is binary. We will point out some properties like the fact that the cooperative game associated it is not a simple game when the target is binary. Additionally, we will briefly comment some comparison between other methods in the literature to determine importances.

Keywords: game theory, cooperative games, data science

GT19.GAMES5
November 10, 2023  12:00 PM
CC2: Conference Room

J. C. Gonçalves Dosantos, R. Martínez, J. Sánchez Soriano

S. Miquel Fernández, H. Norde, H. Hamers, S. el Obadi

N. Llorca Pascual