Abstract
A game theoretic flavoured decision tree is designed for multi-class classification. Node data is split by using a game between sub-nodes that try to minimize their entropy. The splitting parameter is approximated by a naive approach that explores the deviations of players that can improve payoffs by unilateral deviations in order to imitate the behavior of the Nash equilibrium of the game. The potential of the approach is illustrated by comparing its performance with other decision tree-based approaches on a set of synthetic data.
Citare
@Inproceedings{Suciu2022AGT,
author = {M. Suciu and R. Lung},
booktitle = {International Conference on Machine Learning, Optimization, and Data Science},
title = {A Game Theoretic Flavoured Decision Tree for Classification},
year = {2022}
}