The influence of playing styles and team proficiencies on play call predictability in the NFL

National football League (NFL) teams are unusual in team invasion sports as they are an aggregation of sub-teams (offence and defence), hence research conducted on American football should reflect this fundamental difference. However, research has typically quantified technical behaviours related to offensive play. Current understanding is insufficient with respect to offensive and defensive patterns in play given that defensive performance has not been satisfactorily considered. Accordingly, this thesis developed a novel analysis of elite American football stemming from the identification of playing styles. A systematic review of literature revealed that unsupervised dimensionality reduction techniques were predominantly used in team invasion sports (84.21%). Therefore, principal component analysis (PCA) conducted on 5 seasons (2018-2022) of NFL event data, identified 8 distinct styles of play which accounted for exclusively offensive (n = 5) and defensive (n = 3) areas of performance. Subsequent cluster analysis of playing style efficacies demonstrated that teams could be classified into 2-3 proficiency clusters, to varying group membership distributions. Furthermore, multiple linear regression showed that 6 playing style proficiencies were significant predictors of regular season success. Conditional inference tree (CREE) models established the relative importance of 4 playing style proficiencies (passing efficacy, scrambling, rushing and deep passing) in predicting the likelihood of teams executing a pass, run or scramble play. Enhanced by playing styles efficacy among several situational variables, this machine learning approach, to develop a model of playing style efficacy, as a function of team quality, improved the applicability, and most notably, the accuracy, of decision trees as an analytical tool for use in the NFL by approximately 3.35-5.2% respectively for models including and excluding scrambling eventualities.

Level dc:type.qualificationlevel

PhD thesis

Grantor dc:publisher.institution

Middlesex University

Year dc:date.issued

2024

Author dc:creator

• Carter, B.A.

Identifier

oai:repository.mdx.ac.uk:18z55y

OAI identifier oai:identifier

oai:repository.mdx.ac.uk:18z55y