Detecting a Colluding Subset in a Simple Two-Dimensional Game

Abstract

Collusion is covert co-operation between the participants of a game. Detecting the colluding players can require discerning and understanding the player's motivation, which is often difficult task even for humans. In this paper we analyse experimental data from a simple two-dimensional game using synthetic players. We calculate information gains of the features in the data to show how well they indicate collusion. Then we examine J4.8 decision tree classifiers learned from the data and use them to detect the colluding subsets.