Read on Twitter
After a few months of work, my #bdb submission is made public.
I took the approach of modeling players' force-velocity profiles to predict pass outcomes. Represented some unique challenges but I am happy with the outcome.
https://www.kaggle.com/haljordan/modeling-player-biomechanics
I took the approach of modeling players' force-velocity profiles to predict pass outcomes. Represented some unique challenges but I am happy with the outcome.
https://www.kaggle.com/haljordan/modeling-player-biomechanics
Quick overview:
Introduction
The NFL Combine is a display of prospects' strength, speed, and power. For this submission I focused on the 40-yard dash.
Question is: Does the 40-yard dash carry over to what we see in-game? And which qualities transfer to defensive strenghts?
Introduction
The NFL Combine is a display of prospects' strength, speed, and power. For this submission I focused on the 40-yard dash.
Question is: Does the 40-yard dash carry over to what we see in-game? And which qualities transfer to defensive strenghts?
Background
Athlete sprint kinetics are thoroughly researched but do not transfer to field-sports effectively. I tried to figure out a way for individual force-velocity-power profiles to be utilized in-game, while players are in motion.
Lots of math was involved.
Athlete sprint kinetics are thoroughly researched but do not transfer to field-sports effectively. I tried to figure out a way for individual force-velocity-power profiles to be utilized in-game, while players are in motion.
Lots of math was involved.
Results
The model accurately predicted pass outcomes 78% of the time, heavily biasing pass completions. Only 15% of pass incompletions were correctly modeled.
The model accurately predicted pass outcomes 78% of the time, heavily biasing pass completions. Only 15% of pass incompletions were correctly modeled.
Discussion
I found that when defenders were within 0.51s (modeled, not actual) of the receiver, it usually resulted in an incompletion. I adjusted the model to take this into account and it increased accuracy of pass inc0mpletions to 70%, and overall to 88%
I found that when defenders were within 0.51s (modeled, not actual) of the receiver, it usually resulted in an incompletion. I adjusted the model to take this into account and it increased accuracy of pass inc0mpletions to 70%, and overall to 88%
Takeaway
Top speeds are impressive at the NFL Combine but on the field acceleration rates are what separate the good from the great.
I cannot thank @chuurveg enough for his help on the math and @EMsportscience and @ABillowits for ripping my initial draft(s) apart
Top speeds are impressive at the NFL Combine but on the field acceleration rates are what separate the good from the great.
I cannot thank @chuurveg enough for his help on the math and @EMsportscience and @ABillowits for ripping my initial draft(s) apart
