Machine Learning Driven Injury Prediction
summary
Consulting team at QuantumBlack builds a daily risk score to predict lower-body non-contact injuries for an elite football club, training models on consolidated tracking, medical and wellness data reduced to 27 predictive variables despite only ~50 injuries across 3–4 seasons. Heavy hypothesis-driven feature engineering yields sprint-ratio and z‑score features that compare players to themselves and the team, and a LASSO-based stability selection wrapper picks a few robust predictors. A dashboard with explainability shows the model flagged two weeks before a star player’s injury.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 0:22Project Overview
- 0:53Project Start
- 2:59What we did
- 4:36Framing the problem
- 5:37Data Science
- 6:33Project Components
- 10:20Data Landscape
- 12:15Example Hypothesis
- 14:19Aggregation
- 16:04Features
- 16:39Example
- 20:35Challenge
- 21:25Approach
- 22:31Stability Selection
- 26:02Python Implementation
- 27:16Tradeoff
- 27:37Risk Profile
- 28:17Questions
related talks
42mHow Deep Learning is Revolutionizing Sports Analytics
33mUsing machine learning to improve player safety in the NFL
31mBeyond a Bit Fit
24mDeveloping Innovative Systems Through Responsible AI In Healthcare
32mData science vs real world: friends or foes
33mUsing Data Science to Better Evaluate American Football Players
23mEngineering AI-Powered Predictive Healthcare Systems
29mHow the NFL builds computer vision training datasets at scale