2 Bac kgr oun d an d Re lat ed W ork

In hockey, ﬁve players and a goalie per team are on an ice surface and play for a total of 60 minutes. The goal of the game is to put a rubber puck into the opposing team’s net using a 1.5 to 2m long stick made of wood or a composite material. The team who scores the most goals in a game is the winner. In the reg- ular season, if a game is tied after 60 minutes, the teams play an extra 5 minutes of sudden death overtime and after that the game is decided by a shootout. In the playoﬀs, after 60 minutes, additional 20 minute overtime periods are played until a team scores. As far as the authors are aware, there is no previous work, in the machine learning community, to predict the winner in a hockey game. Machine learning has been used in other major sports with a varying degree of success to predict the outcome of games, championships and tournaments. Most of the researchers employed neural networks for this task. Chen et al. [1] were among the ﬁrst to use neural networks for making predictions in sports. They used neural networks to make predictions in greyhound racing and their classiﬁer was shown to be able to make a proﬁt. Huang and Chang [2] used neural networks to make predictions of game winners in the 2006 World Cup and was able to achieve an accuracy of 75%. Purucker [3] used neural networks to make predictions in the National Football League using only four categories he was able to make prediction accuracy of 78.6%. Pardee [4] used neural networks to make predictions for the outcome of the NCAA football bowl games and returned a similar accuracy of 76%. Loeﬀelholz et al. [5] use neural networks to predict outcomes in the National Basketball Association (NBA) and using common statistics found in the box score of NBA games his model was able to predict with 74.33% accuracy. While neural networks are primarily used in literature, authors have mentioned the use of other classiﬁers; however, these have not worked as well as neural networks such as [6].

3 Nat ion al H ock ey L eag ue C ase Stu dy