Construction of the Winner Predictive Model in Major League Baseball Games: Use of the Artificial Neural Networks

Abstract

The purpose of this study was to construct the winner predictive model in Major League Baseball (MLB) by using the artificial neural networks (ANNs) as a tool. One hundred and twenty-six games from the seasons of 2006 to 2012 held between New York Yankees and Boston Red Sox teams were involved in this study, and 60 variables were collected to analyze. Meanwhile, the winner predictive model was applied to the season of 2013 games to verify the model accuracy by virtual bets. In addition to ANNs, logistic regression was also used to develop the winner predictive model to discuss the differences, and to explore the advantages and disadvantages between these two tools. The results showed that ANNs-based predictive model had 72.22% of winning accuracy. Moreover, for the season of 2013 games between Yankees and Red Sox teams were used by way of virtual bets to test the model's accuracy, and the results showed 73.68% of accuracy. Because of combined many areas, such as sports, sports industry, management and quantitative methods, and also equipped the ability of data mining, our findings provided an accurate, simple and reasonable way to predict the winning team for bettors. Therefore, the use of ANNs model was recommended as a suitable tool to predict the winning team.