Project outline
Developed a new metric (PoQIT: Pocket Quality In Time) to analyze the pocket established by the offensive line during a play using tracking data provided by the NFL.
Combined multiple sources of data including highly-granular, complex tracking data during the ETL process.
Completed Feature Engineering to create a series of new features that capture some more complexity of football.
Trained both XGBOOST (eXtreme Gradient Boosting) and BART (Bayesian Additive Regression Trees) models on the PoQIT metric.
Conducted a Network Shortest-Path Analysis to see which positions had the shortest Time to Close Distance between two nodes.
Compared the Feature Importance from both models to develop a list of prescriptive steps for an NFL team to enact.
Project Summary
PoQIT Visualizations: The visualization of the PoQIT is describing the value attributed to the blockers during the course of the down. It is measuring the effective area each blocker is creating, with more value earlier in the down. Green coloring is higher value of PoQIT, red means that this is a lower PoQIT value.
Modeling Values: The weighted area per person over time is a graph showing the weighting changing over the course of the down to give more value to earlier in the play, as earlier pressure is more disruptive to the play. The following plot is showing the outputs of the model on a week-level aggregation. This is simply to show that the values are roughly normal in shape, which means that our data is not being skewed by outlier plays in one direction or another.
Feature Importance: The two feature importance graphs show which inputs to the model were the most important towards computing the PoQIT metric. This highlights some prescriptive steps for an NFL team, because these features are what causes a good PoQIT or from a defense's perspective what can be done to disrupt a PoQIT. Some of the top features include play-action, motion at the snap, number of rushers, and speed of the Defensive line.
Top Teams: Finally, we have a table of the top and bottom 5 teams in terms of the defense's perspective along with some relevant metrics like defensive line speed and average number of rushers.