Predicting the results of NFL games using machine learning

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
School of Business | Master's thesis
Degree programme
Information and Service Management (ISM)
42 + 7
The NFL is the most popular sport in terms of viewership among the leagues in the United States, and is an entertainment business worth billions of dollars. Being able to predict the outcome of games is a popular topic among bettors, academics, but also among analysts who want to determine which game events most influence games in the long run. Additionally, predictions can serve as infotainment to provide context to spectators, and machine learning models can be utilized for other beneficial aspects such as injury prevention among athletes. This thesis aims to take data from the most recent NFL seasons to find out if machine learning models can be built to accurately predict the outcome of games, and additionally see which statistics and in-game events weigh most heavily on the outcomes. The literature review of the thesis covers previous research of machine learning models in American football, and also in other professional sports. The data is sourced from Pro Football Focus’ website and includes the scores and team statistics from two recent NFL seasons. Three different machine learning models are applied on the dataset to compare their performances and results. All three models reached quite high accuracy scores, with the lowest model having 85% accuracy in predictions. In conclusion, it seems possible to apply machine learning to predict the outcome of NFL games. There is room for further research by taking more seasons into account, or possibly comparing the results of the models on a dataset of college American football games for example.
Thesis advisor
Malo, Pekka
machine learning, NFL, sports, prediction model, logistic regression, random forest classifier, support vector machines