Proposal of an automatic tool for evaluating the quality of decision-making on Checkers player agents

Abstract

Checkers player agents represent an appropriate case study for the best unsupervised methods of Machine Learning. This work presents a tool to measure the performance of these methods based on the quality of the decision making of these agents. The proposed tool, based on the data of movements performed in real games by the agents under evaluation, provides a statistical way of automatically comparing the coincidence rates between the decision making of the evaluated agents with those that the remarkable player agent Cake would do in the same situations. The tool was validated through tournaments between agents comparing their respective coincidence rates and their performance.