Overview

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 bridges raw 10 Hz tracking data and football-relevant insights through feature engineering, metric design, and model-based evaluation.

---

Analytical Approach

Traditional receiving metrics (yards, targets, catches) describe outcomes but miss the process. With player-tracking data, this project models how a play unfolds in space and time:

• Post-throw trajectories — Receiver movement paths after the ball is released
• Spatial convergence — How quickly and efficiently the receiver closes distance to the ball landing location
• Separation dynamics — Defensive proximity over time, measured as continuous signals rather than binary outcomes
• Contextual factors — Route type, coverage scheme, and game situation as conditioning variables

---

Feature Engineering & Modeling

Extensive feature engineering is applied to raw tracking data:

• Receiver and defender positions, velocities, and directions at each frame
• Relative distances and angles between players and ball
• Time-aligned features from throw to catch/incompletion
• Normalization of play direction for consistency

The project explores LSTM sequence models for post-throw movement prediction, with baseline regression and tree-based comparisons. Model residuals are used to derive performance metrics that isolate player contribution beyond expectation.

---

Derived Metrics

A key contribution is the construction of continuous performance signals:

• Expected vs actual spatial convergence — How well did the receiver track the ball compared to model expectation?
• Separation quality over time — A time-series metric rather than a single snapshot
• Receiver efficiency — Context-adjusted, comparable across players, and interpretable for coaches and analysts

---

Tech Stack

• Python, Pandas, NumPy, Scikit-learn
• LSTM (sequence modeling)
• Matplotlib, Plotly, Jupyter Notebooks
• NFL Big Data Bowl player-tracking dataset (10 Hz)