Abstract
Decision trees and random forests are some of the most popular machine learning tools for binary classification, being used in many practical applications. Both methods provide a neighborhood for tested data during the prediction phase, and probabilities are usually computed based on the proportion of classes in those neighborhoods. The approach presented in this paper proposes replacing the prediction mechanism with one based on a probabilistic classifier based on the Nash equilibrium concept applied to the local data selected by the random forest classifier. Numerical experiments performed on synthetic data illustrate the behavior of the approach in a variety of settings.
Citare
@Inproceedings{Suciu2022ANG,
author = {M. Suciu and R. Lung},
booktitle = {Hybrid Artificial Intelligence Systems},
title = {A New Game Theoretic Based Random Forest for Binary Classification},
year = {2022}
}