A Machine Learning Based Method for Rating Adversarial Board Games

Blockchain-based games that enable players to engage in peer-to-peer item transactions have recently gained remarkable popularity. Since these games transfer in-game assets based on match results, appropriate player matching according to their skill levels is becoming more important. However, the traditional Elo rating system, which is widely used to rank player abilities, faces significant challenges. One primary problem is "smurfing," where experienced players create new accounts to play against less skilled opponents. This paper proposes a new rating system called ML_Elo that is tailored for the game of Othello. By harnessing the power of machine learning, this system aims to more accurately predict a player’s skill and appropriately adjust the fluctuation value in his/her rating. This promotes fairer in-game player matching and mitigates the negative impacts of smurfing and other gaming malpractices. The proposed method is applied to a virtual Othello environment using AI players and search for optimal parameters. The creation of the AI players utilizes a learning model based on neural networks trained on actual game records. The experiment evaluates the transition of individual players’ rating values and the distribution of overall rating values, comparing the proposed method with the standard method.