Poisson Model and Bradley terry Model for predicting multiplayer online battle games

There has been increasing interest for predicting game results both in lottery industry and in academia, however there have been few literature for predicting Aeon Of Strife (AOS) type game match results. Moreover, few studies have been conducted for comparing different prediction models for game matches. In this paper, we compare Poisson Model and Bradley Terry Model for League of Legends (LOL) matches from 2013 to 2014 in South Korea. For Poisson model, we adopt time dependent bivariate Poisson regression model proposed by Dixon and Coles. For Bradley Terry Model, we add Davidson method to allow tie count. From the constructed models, we estimate maximum likelihood values, and present performance evaluation results on the training data. The performance evaluation results indicate that the adopted models in this paper are effective in prediction of actual LOL match results.