American football is played primarily as a sequence of two types of plays - runs and passes, determined before the play by the team that has the possession of the ball. All other plays are much less frequent, and considered "special plays". In this study we show that the type of play can be predicted using pattern recognition methods with accuracy higher than random chance based on several indicators that reflect the status of the game such as the down, time left in the game, score difference, etc. These values were used to predict the next play by using Support Vector Machine and Weighted Nearest Distance classification schemes. Experiments with data from all plays in 11 National Football League (NFL) seasons show that the ability to predict the next offensive play can be as high as 74% for an entire season of a single team.
© Copyright 2014 International Journal of Computer Science in Sport. Sciendo. Kaikki oikeudet pidätetään. / Kääntänyt https://www.DeepL.com/Translator