Formulating Major League Baseball Playoff Prediction as a Ranking Problem

Among all professional sports, the major league baseball (MLB) in the United States is the first that embraced extensive data collection for player and team performance evaluation and analysis. The practice has been recognized as sabermetrics. Indeed, MLB has the most abundant data available for performing sports analytics. In MLB, data-driven analysis has been carried out to predict various aspects of the game, ranging from player-specific characteristics (such as performance, valuation, contract renewal) to team-specific performance (single-game outcome, playoff probability, and championship probability). In this paper, we focus on the prediction of playoff probability in MLB using publicly available team statistics. Unlike the popular approach to making binary classification for each team regarding whether or not the team can make the playoff, we propose to formulate the playoff prediction as a ranking problem. We argue that formulating playoff prediction as a ranking problem is a much better fit for the MLB playoff team selection rules. We also show that our playoff prediction is significantly better than those reported in the literature and online articles.