03 — Data Analysis

NFL Draft Prediction

Predicting Draft Outcomes for College Quarterbacks — COGS 108

RoleData Analyst (Group)

CourseCOGS 108 — Data Science in Practice

ToolsPython · Pandas · Matplotlib · Seaborn

Research Question

Predicting draft outcomes from combine data

This COGS 108 group project asked whether a college quarterback's combine numbers and college statistics can predict whether they get drafted.

Using NFL Combine and NCAA datasets, we tested whether measurables like 40-yard dash time, TD/INT ratio, and total yards separate drafted from undrafted quarterbacks.

Trends exist, but no single metric is decisive. Draft outcomes depend on performance, athleticism, and context together, which changes how scouting data should be presented.

View on GitHub →

MyFinalProject_SP25 Public

📓ProjectProposal_Group072_SP25.ipynb

📓DataCheckpoint_Group072_SP25.ipynb

📓EDACheckpoint_Group072_SP25.ipynb

📓FinalProject_Group072_SP25.ipynb

Jupyter Notebook — 100%

40-yard dash time by draft verdict — actual project output

Boxplot: 40-yard dash time by draft verdict, from the actual analysis

TD/INT ratio by draft verdict — actual project output

Boxplot: TD to interception ratio by draft verdict, from the actual analysis

Methodology

How the analysis was run

The project followed a standard EDA process: collecting and cleaning data from two sources, visualizing the distributions, running correlation analysis across the key metrics, and finishing with a hypothesis test to check whether the differences were statistically meaningful.

The metrics analyzed were 40-yard dash time, vertical jump, TD/INT ratio, completion percentage, and total passing yards. None of them cleanly separated the two groups.

Finding

What the data showed

Drafted quarterbacks post better combine numbers and college statistics on average, but the two groups overlap heavily. Quarterbacks with below-average measurables get drafted, and quarterbacks with strong numbers go undrafted.

For design, that means a scouting tool should show distributions and confidence ranges rather than single-metric rankings. A bar chart of the top 40-yard dash times would misrepresent the data.