This project explores advanced player-tracking analytics using the NFL Big Data Bowl dataset, with a focus on wide receiver movement after the throw, spatial convergence toward the ball, and contextual performance beyond traditional box-score metrics.
The work emphasizes feature engineering, metric design, and model-based evaluation of receiver behavior during pass plays, bridging raw tracking data and football-relevant insights.
Traditional receiving metrics such as yards, targets, and catches describe outcomes, but they do not fully capture:
- How well a receiver moves toward the ball
- How quickly separation is created or maintained
- How receiver performance varies by route, coverage, and game context
With player-tracking data, we can analyze how a play unfolds in space and time, rather than only how it ends. This project focuses on modeling and measuring those dynamics in a way that is interpretable and relevant to football decision-making.
The analysis centers on post-throw behavior, specifically:
- Receiver movement trajectories after the ball is released
- Spatial convergence between receiver and ball landing location
- Defensive proximity and separation dynamics
- Contextual factors such as route type and coverage
Rather than treating plays as static events, the project models them as continuous spatiotemporal processes.
Extensive feature engineering is applied to the raw tracking data, including:
- Receiver and defender positions, velocities, and directions
- Relative distances and angles between players and ball
- Time-aligned features from throw to catch/incompletion
- Normalization of play direction for consistency
These features form the basis for downstream modeling and metric construction.
The project explores predictive and descriptive modeling approaches, including:
- Sequence-based models for post-throw movement
- Baseline regression and tree-based comparisons
- Residual-based metrics to isolate player contribution beyond expectation
Model outputs are used not only for prediction, but as inputs to derived performance metrics.
A key objective is to move beyond binary outcomes (catch/no catch) and derive continuous performance signals, such as:
- Expected vs actual spatial convergence
- Separation quality over time
- Receiver efficiency relative to modeled expectation
These metrics are designed to be:
- Context-aware
- Comparable across players
- Interpretable for analysts and coaches
- Source: NFL Big Data Bowl player-tracking dataset
- Granularity: 10 Hz player and ball tracking
- Scope: Passing plays with receiver targets
- Supplementary data: Player metadata, play context, routes
Figure: metric correlation view of derived receiver metrics with contextual features.
There is a positive correlation between TrueSpeed and Expected Points Added (EPA), but a negative correlation between Route Execution and EPA.
Big_Data_Bowl/
├── models/ # LSTM outputs
├── notebooks/ # Feature engineering, modeling, and analysis
├── outputs/ # Figures and analytical outputs
└── README.md
This framework enables analysts to:
- Evaluate receiver performance independent of catch outcome
- Compare players with similar box-score stats but different movement quality
- Understand how separation evolves post-throw
- Identify route and coverage contexts that amplify or suppress receiver effectiveness
- Post-throw receiver movement analysis
- Feature-driven modeling of spatial dynamics
- Metric construction from model residuals
- Broader route- and coverage-specific analysis
- Integration with team-level offensive tendencies
- Extension to multi-season comparisons
- Refinement of metrics for scouting and coaching applications
- Python
- Pandas, NumPy
- Scikit-learn
- Matplotlib, Plotly
- Jupyter Notebooks
This repository represents a research-driven exploration of player-tracking analytics developed in the context of the NFL Big Data Bowl. The emphasis is on translating complex spatiotemporal data into actionable football insights through principled feature engineering, modeling, and metric design.