A combination of two of my favorite interests: analytics and college football. My college football prediction model uses a sports data API to draw team stats from every college football game each week into an excel database. Each team is assigned a set of weighted stats based on their performance in past games adjusted for the strength of their opponent. Using these stats for any two teams, the model predicts game performance metrics such as yards per rush, total yards, and scoring efficiency. These stats are used to calculate a predicted score. This final score is then adjusted using a correction factor calculated from how accurately the model has predicted each of the two teams' past games.

 

 

 

 

 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

 

After receiving adequate game data (~4 games for each team), my model achieved a success rate of above 55% against the Vegas Spread most weeks. I believe the model will perform even more accurately in future seasons, as, by design, it is unable to account for a lack of inter-conference play and the corresponding lack of common opponents. For example, the power ratings I generated using a MATLAB script place BYU as the 3rd strongest team, likely due to dominating its relatively weak schedule. In a normal, non-covid season, BYU would have common opponents with Power Five teams and would therefore receive less credit for dominating weaker teams.