Machine LearningApril 14, 20242 min read

NFL Big Data Bowl

Problem Statement

What separates a successful tackle from a missed one? Using Kaggle's NFL Next Gen Stats dataset from the 2024 Big Data Bowl, this analysis identifies the physical and situational factors most predictive of tackle success. The dataset captures player tracking data at sub-second resolution: speed, acceleration, angle of pursuit, and field position for every play.

Methodology

Feature Screening

Each tracking metric was tested independently against tackle outcome using logistic regression to identify the strongest individual predictors before building a multivariate model.

Feature importance showing tackle acceleration as the strongest predictor — Positive and negative predictors of tackle success. Tackle acceleration and pass-caught events are the strongest positive predictors.

Results

Position Analysis

Some defensive positions are significantly more likely to make successful tackles than others.

Tackle success rate by defensive position — Defensive positions ranked by tackle success probability.

Offensive Formation Impact

Tackle success is highest when the offensive formation is empty, and lowest when the formation is singleback.

Tackle success rate by offensive formation — Tackle success rate broken down by offensive formation type.

Speed & Acceleration

As expected, tackle speed and tackle acceleration are correlated with the probability of a successful tackle.

Missed Tackle Locations

Heatmap showing missed tackles occur most frequently in the middle of the field — Missed tackles more commonly occur in the middle of the field.

Next Steps

Examine prior years' NFL Next Gen Stats Data for trend analysis
Compare individual player tackle likelihoods and their impact on team winning probability
Build a model that predicts total tackles per player per game or season