experience, fostering strategic decision-making and competitive gameplay.
2.2 NFL Data Science
The NFL players statistics are collected based on position and how the players
perform against other teams, whether they are offensive players, defensive players, or
special teams players on a weekly and yearly basis. Quarterback statistics are collected
based on pass attempts, pass completion, yards (air or rushing), touchdowns (passing
or rushing), interceptions, and fumbles (total amount and amount lost). Running back
statistics are collected based on rushing attempts, yards (rushing or receiving),
touchdowns (rushing or receiving), and fumbles. Wide receiver and tight end statistics
are collected based on catches, yards, touchdowns, and fumbles. Whereas defense
statistics are collected by team and individual players, fantasy football is collected only
by team. In fantasy football, this statistic consists of yards allowed, touchdowns
(allowed and returned), sacks, punt returns, kickoff returns, punt blocks, field goal
blocks, and turnovers. Kicker statistics are collected based on field goals made, field
goal attempts, yard ranges kicked, and blocked field goals.
NFL statistics are very important because it determines the outcome of a
player and team performance. There are many factors that come into performances such
as the team’s coaches, training staff, and injuries. NFL statistics can outline a player or
team strengths and weaknesses (Khanacademy, n.d.). For example, a team’s defense
can be ranked 32nd for allowing the most amount of yards per game. It allows teams to
have a comprehensive view of their opponents.
2.3 Feature Selection
In constructing our predictive models, we underwent a thorough process of
feature selection to encapsulate the intricate dynamics of player performance in fantasy
football. Drawing from a rich dataset comprising fifty distinct statistics for each player,
our models embraced an extensive array of season-to-date features. The term "season-
to-date" signifies the computation of a rolling average for each statistic up to, but
excluding, the current week's game. This methodology facilitated the incorporation of
past performance metrics as predictors of future player performance. Noteworthy
statistics integrated into the model encompass passing yards, passing touchdowns,
interceptions thrown, rushing yards, receptions, and fumbles lost. (Porter, Predictive
Analytics for Fantasy Football: Predicting Player Performance Across the NFL 2018)
In tandem with season-to-date features, we incorporated game-specific characteristics
to augment predictive accuracy. Binary indicator variables were devised to capture
pivotal attributes of each game, including whether the player competed at home or
away. We recognized the potential influence of home-field advantage on player
performance, thus deeming it essential for inclusion in our models. (Porter, Predictive
Analytics for Fantasy Football: Predicting Player Performance Across the NFL 2018)
Moreover, dummy variables were formulated to account for the offensive inclination
of the player's team, acknowledging that team strategies and dynamics can significantly
impact individual player performance. Lastly, dummy variables for the player's
opponent were integrated into the model, reflecting the caliber of the opposing team,
which in turn can influence player scoring potential. (Baak, Fantasy Football
SMU Data Science Review, Vol. 8 [2024], No. 2, Art. 7