Overview
Our project centers on an in-depth analysis of the NFL, ranging from salary cap dynamics to draft value assessments, player statistics, and team performance metrics. Through rigorous data exploration and advanced analytical techniques, we aim to unearth insights that inform strategic decision-making for NFL teams, particularly in resource allocation and talent acquisition.
We focus on how different forms of resource allocation — salary cap space and draft picks into certain positions — correlate to individual player and team success. We plan to answer questions like: should a team invest in younger, cheaper quarterback prospects with draft capital, or go after proven veteran winners with significant cap space? After analyzing resource allocation and positional value, we create effective data visualizations, then build an ML model predicting team success for the upcoming season from that data.
Training
Project members will need to know:
- Basic NFL knowledge — can be taught in a week or two: how the salary cap works, how draft capital works (draft value charts), positional terminology, and basic familiarity with advanced NFL analytics such as PFF grades, WAR (wins above replacement), AV (approximate value), and EPA (expected points added)
- Data science and manipulation tools — Pandas, NumPy, SciPy
- Web scraping tools — Beautiful Soup or Selenium
- ML tools — scikit-learn (decision trees, CNNs); other libraries welcome
- Data visualization tools — Matplotlib, Seaborn
Milestones
September 15 — Web scrape and gather data
Three teams form: one scrapes basic salary cap and draft data, one gathers player evaluation data based on a preferred metric (PFF, AV, etc.), and one gathers team evaluation data (PFF, team website, winning percentage). Once scraped into CSVs, each team normalizes the data with Pandas — normalizing positional terminology to an agreed standard, converting raw numbers to percentages where possible, and grouping data by team or year.
September 30 — Data visualizations
Team 1 takes salary cap, draft, and player data to analyze player return on investment. Team 2 takes player performance and team success data to analyze positional value. Team 3 combines player ROI and positional value data to generate team success insights based on positional investment.
October 15 — Develop the ML model
Predict team success for next season from new and pre-existing draft and salary cap data. Team 1 creates a model with team investment as features and player success metrics as output; Team 2 creates a model with player success as features and team success metrics as output. Model types include RNNs, time series (ARIMA), and decision trees. Two weeks per model type, then two weeks to combine them so the labels of the first become the features of the second (team success → predicted player → predicted team).
December 15 — Report and poster
Develop visualizations for model performance and accuracy, and prepare a report and poster acknowledging findings.