Once again, the goal of this project was to aim to create a model that could predict whether the total number of points scored in an NFL game would go over or under the total line with an accuracy of over 52.4%. This had to be done with only statistics and information that would be available prior to the game starting. This is because this project focuses on predicting the outcome of the total line before the game begins. There is live betting that can be done during games, but that would require a whole different set of data and a different approach than this project took.

Just a quick reminder on why this project is important. 99 billion dollars (www.businesswire.com) was wagered on sports in the past year (2022). That is a massive number and it’s expected to keep growing at an annual rate of 9.8%. The past few years have seen tremendous growth in the sports betting industry. This is largely due to sports betting on pro and college sports being made legal at the federal level in 2019. However, this had to be put into effect on a state by state basis. Currently there are 36 states that have legalized sports betting. Although the specifics differ by state, like Washington where online sports betting is illegal, but in person sports betting at tribal casinos is legal (www.forbes.com). The passing of this law prompted easy access to sports betting at a level never seen before. In person sportsbooks were being created in states other than Nevada for the first time. What really changed the industry was the creation of online sports betting platforms like DraftKings and Fanduel. This allowed anyone to sports bet from their phone or other device without ever having to leave their house. This had to approximately 1 in 5 Americans betting on sports in the past 12 months, with approximately 50% of Americans saying they’re interested in sports betting (www.pewresearch.org). With this growth in popularity many new sports bettors or people interested in sports betting likely have the question: Is it easy to be a profitable sports bettor? The simple answer is no. Assuming bets are only placed on totals and spreads with the typical -110 odds, a win rate of 52.4% is needed to be profitable. This doesn’t sound too difficult. However, sports books are very good at setting their lines to make it difficult on the bettor. To put it into perspective professional sports bettors will consider a 55% win rate a good year (www.actionnetwork.com). So, this project will attempt to create a model that will help sports bettors, especially new sports bettors, stay profitable in a rapidly growing industry.

Finally, before the overall results of the project are revealed, it is important to touch on the difficulties that this project faced at becoming successful. Coming in, this project was not easy or guaranteed to be successful. Some quick background research and general knowledge of the sports betting field, reveals that creating a model that will correctly predict the total result more than 52.4% of the time will be challenging. There are a few reasons for this. First, finding useful NFL data that also contains sports betting data or can be combined with sports betting data is not an easy task. Second, there are just not that many strong predictors of the total result, and the ones that are strong tend to be uncommon. Third, sports books have been perfecting their models for setting total lines for a long time. As a result, they are very effective and difficult to beat.

So, was this project a success? Yes and no. The model accuracy results are below.

Model	Accuracy
Decision Trees	0.53
Naive Bayes	0.52
Logistic Regression	0.52
Support Vector Machines	0.54
Random Forest	0.52
XGBoost	0.51
Neural Network	0.5438

Three of the models had an accuracy above the 52.4% cutoff point necessary to be profitable. That initially seems like it should be deemed a success. However, only predicting with accuracy ~2% higher than necessary leaves a very small margin for error. Also, it will only generate meaningful profits if a sports bettor is placing somewhat large wagers every time. For example, if a sports bettor placed $100 bets on every game of the upcoming NFL season (272 games), the sports bettor would profit $1051.50. However, if a sports bettor placed $5 bets on every game of the upcoming NFL season (272 games), the sports bettor would only profit $52.57.

Amount Bet: 100

Decimal Odds for -110: 1 – (100 / Sports Odds) = 1 – (100 / -110) = 1.91

Total Amount Won Per Bet: Amount Bet * Decimal Odds = 100 * (1.91) = 191

Profit Per Bet: Total Amount Won Per Bet – Amount Bet = 191 – 100 = 91

Profit Over 272 Games Betting $100: (91 * (272 * 0.5438) – (100 * (272 * 0.4562) = 1051.50

Amount Bet: 5

Decimal Odds for -110: 1 – (100 / Sports Odds) = 1 – (100 / -110) = 1.91

Total Amount Won Per Bet: Amount Bet * Decimal Odds = 5 * (1.91) = 9.55

Profit Per Bet: Total Amount Won Per Bet – Amount Bet = 9.55 – 5 = 4.55

Profit Over 272 Games Betting $5: (4.55 * (272 * 0.5438) – (5 * (272 * 0.4562) = 52.57

So, determining if this project was actually successful, comes down to the goals of each individual sports bettor. If the goal is just to sports bet as a hobby and just break even or make a small profit, then this project would be deemed a success. However, if the goal is to be as profitable as possible, then this project wouldn’t be considered very successful as it requires large bet sizes to see meaningful profits. This project was challenging and the overall outcome was not explicitly what was desired. However, it did provide plenty of opportunities to practice machine learning and learn new skills, especially what to do when issues arise. There are a few areas that can be further done in the future for this project. A CNN model could be attempted. CNN’s are generally used for image data. So, it’s unlikely it would result in an improved accuracy, but it’s still something that could be tried. Apart from that, the only other way to improve upon the accuracy would be finding more predictive features of the total result for NFL games.