Below, are questions that could potentially be answered through this project.

• What variables best predict an NFL game going over or under the total?
  • The most important variables appear to be wind : the wind speed, temp: the temperature, surface: the type of playing surface, total_line: the total line for the game, andtotal_qb_elo : the combined elo rating for the two starting quarterbacks (essentially tells us how good the quarterbacks are). However, to confidently determine which variables are most important, actual variable importance methods would need to be performed.
• What variables best predict, which team will cover the spread in an NFL game?
  • This project initially focuses on the total, not the spread. This will be addressed in the future.
• What is the general sentiment around sports betting in the news today?
  • The general sentiment is mostly positive. The number of people placing sports bets is increasing, large companies (DraftKings, FanDuel, Fanatics) are finding revenue opportunities in sports betting, and mobile sports betting is making it more accessible. All these things contribute to a positive sentiment. The main negative sentiment is generally coming from Native American tribes. Before, sports betting, especially mobile sports betting, started to become legal, the main place sports betting was allowed was on Native American reservations. So, Native American tribes, see the widespread legality of sports betting as a threat to their profits.
• Has the general sentiment around sports betting in the news improved, gotten worse or stayed the same compared to 20 years ago?
  • The sentiment has definitely improved. This is most likely due to the widespread nature of sports betting becoming legal in many states. States and larger companies are also seeing this as a revenue opportunity. These two things lead to more positive news around the topic of sports betting being released.
• Does a coaches 4th down decision making have a significant effect on a total in an NFL game?
  • The data necessary to answer this question was not obtained, due to a lack of time. This is also a question that could potentially be answered in the future.
• Which type of machine learning classification model will best predict whether an NFL game will go over or under the projected total?
  • Decision Trees: 51.3%
  • Naive Bayes: 52.7%
  • Support Vector Machines: 53.5%
• Are home teams more likely to cover the spread compared to away teams?
  • This project initially focuses on the total, not the spread. This will be addressed in the future.
• Are games played in domes more likely to go over the total than games played in outdoor stadiums?
  • 51% of games played in covered stadiums went over
  • 49% of games played in covered stadiums went under
  • 46% of games played in non-covered stadiums went over
  • 54% of games played in non-covered stadiums went under
• Can a model be generated that can correctly predict the total or spread at least 52.4% of the time?
  • Yes, but with the current data and variables available, the model only slightly outperforms the 52.4% accuracy necessary to be profitable. The Support Vector Machine model had an accuracy of 53.5%. However, only predicting with an accuracy 1.1% higher than necessary leaves a very small margin for error and won’t generate meaningful profits unless the sports bettor is placing very large wagers every time.
• Are playoff games more likely to go under the total compared to regular season games?
  • No, there is no significant statistical difference for playoff games going under the total line compared to regular season games.
• Do certain referees make an over or under more likely to occur?
  • Project did not produce an answer to this question. It would be possible to find out this question with the data used in this project in the future.