• 

  Live Gameday

• 

  Game Signup

• 

  ML Prediction Output

NextDown

Approved for Beginner Track

Inspiration

High school football coaches often have limited coaching staff and resources, yet they must make play calling decisions in about 40 seconds between plays. Our entire team experienced this firsthand in our high school athletic program where one coach may handle multiple responsibilities during a game.

We built NextDown to help high school coaches process game data quickly and make smarter play calls using machine learning.

What it does

NextDown is a mobile web app that analyzes game context, defensive formations, team tendencies, and weather conditions to recommend the top three plays and formations most likely to succeed.

How we built it

We trained a machine learning model using football play by play data and contextual features such as down, distance, field position, defensive alignment, score, and weather from NFL_Data_Py. Environmental data is pulled using the Open Metio API.

The system evaluates multiple play categories and ranks them by predicted success probability.

Challenges

One challenge was balancing model accuracy with limited development time. Our early model achieved roughly a 20% match rate, meaning the predicted plays matched successful outcomes about 20% of the time. We also had to decide which data should be entered pre-game, per drive, or per play to keep the interface simple for coaches. We roughly finished with an 50% match rate to an NFL coach

What we learned

This project taught us how to combine sports analytics, machine learning, and product design into a tool that could realistically assist coaches with limited resources.

What's next

Future improvements include improving model accuracy with more data, integrating real-time tracking, and expanding the system to adapt to live game adjustments.

Built With

• numpy
• open-meteo
• pandas
• pickle
• python
• react
• scikit-learn
• vite