NFL Edition
The 2022–2023 NFL season has officially begun! Players are fresh and locked in, fantasy football rosters are set, Analyst predicting subjective results, tailgating supplies ready to go, money to be made(or lost) with some fun bets. There is nothing like this time of the year. As fresh new Data Scientist in IBM‘s Client Engineering Team, I would like to add some data science, machine learning, and artificial intelligence to the fun. Utilizing IBM technology Cloud Pak for Data and 20 years of NFL data, I created a different way to analyze all 32 NFL team’s game performances!
Each week for the upcoming NFL season, every game will be entered into the AI Performance Analyzer (APA). APA will analyze the team’s performance (majority offensive stats) and determine if they played a solid game statistically to win the game despite the actual outcome. Both team’s performance will be fed to APA separately with key information unknown, the team and score.
Team and Score Unknown to AI Athletics Performance Analyzer.
Why wouldn’t APA see the team you ask? If APA knows each team it will then only compare the team to itself not the entire NFL. For example, a Saints game performance will be measured against 20 years of Saints game performances instead of the entire 20 years NFL game performances. Although, APA learning games performances with the team would be a good analysis for another day.
Score unknown also? Yes its unknown, it was a game statistic I went back and fourth with about showing APA. We all know the score ultimately decides the game, but does it really say much about performances. Low scoring games can be good games performances statically but the reason for low score could be 80 yard drives ending in field goals. I believe high scoring games could overshadow the low scoring games influencing APA to label low scoring games as losses. Very tough decision to make but for this year score will be unknown, but in next iteration score could make an appearance.
Game Performance Statistics Consumed by APA
First Downs, Third Down Conversions, Fourth Down Conversions, Passing Yards, Rushing Yards, Total Yards, Completion Percentage, Sacks(Allowed), Rushing Attempts, Fumbles(by offense), Interceptions(by offense), Penalties, Redzone Conversions, Drives, Defense/Special Teams TouchDowns, Time of Posses-ion(minutes), Penalty Yards, Sack Yards.
We want APA to only see the stats above for each team in each game for better results and evaluation of game performances. The APA results will be revealed for each team then compared to their opponent for the Tough Game Analysis.
The APA results for each game will be put to the test in the Tough Game Analysis comparing the actual game outcomes. The Tough Game Analysis will label and score from each teams point of view in each game using the APA results in the following way.
Top Level Results
Tough Games – APA results reveal that the opponent’s game performance was enough to win the game.
Easy Games – APA results reveal that the opponent’s game performance was not enough to win the game.
In-Depth Results (actual game outcomes vs APA results)
TGA in-depth analysis looks at both results, actual and APA, labels them and scores on a difficulty level.
I’m expecting to see really interesting results such as games where APA reveals both teams to win, both teams to lose, or an opposite result where the team revealed to win but lost and their opponent revealed to lose but of course won.
Difficulty Score
The TGA’s Difficulty Score will be used to rank each at the end of each week. These scores will accumulate throughout the season revealing which team played in the toughest games and which had an easy road.
Weekly post sharing and breaking down the results by team, game, and conference. I will also add my thoughts and opinions about if Artificial Intelligence got it right. Open discussion welcomed.
The creation of AI Athletics Performance Analyzer (Technical Write Up)
Improved results and analysis with more IBM technology. IBM Watson’s Openscale, AutoAi, and Cognos.
