NFL Wide Receiver Blocking Analytics

This repository contains the validation component of a six-person Bruin Sports Analytics research project on wide receiver blocking effectiveness. The full paper develops a quantitative framework from play-level and player-tracking features; this notebook evaluates whether an interpretable decision tree can separate above- and below-median blockers under the project's volume-adjusted Blocking Effectiveness Score (BES).

Result

Using a stratified 80/20 split and a depth-limited decision tree, the validation notebook produced 84.13% test accuracy on the project dataset. The checked-in notebook preserves the confusion matrix, classification report, and learned rules from that run.

My contribution

I was responsible for the decision-tree validation: selecting the validation features, defining the above-median classification target, evaluating the model, and producing an interpretable tree for the team.

Repository contents

• BSAFINAL.ipynb - decision-tree validation and visualization.
• A Quantitative Framework for Assessing Wide Receiver Blocking Effectiveness Using Player Tracking Data.pdf
  • final research paper.

Reproducing the validation

The source player-tracking dataset is not redistributed in this repository. Provide the prepared final_BES.csv file separately, then:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
mkdir -p data
cp /path/to/final_BES.csv data/final_BES.csv
jupyter notebook BSAFINAL.ipynb

To use another location, set BES_DATA_PATH before running the notebook:

export BES_DATA_PATH=/path/to/final_BES.csv

The generated tree image is written to artifacts/bes_tree.png by default.

Method summary

The validation uses four features:

• median yards generated after a block;
• normalized blocking volume;
• median skill-player downfield BES;
• yards generated per blocking frame.

The target is whether a player's volume-adjusted BES is above the dataset median. The classifier uses max_depth=4, min_samples_leaf=5, balanced class weights, and a fixed random seed.

Data and attribution

This was collaborative research conducted through Bruin Sports Analytics. The paper in this repository contains the complete methodology, authorship, and research context. The notebook here represents my decision-tree validation contribution rather than the full team pipeline.