Smart Advisor Substitution

Inspiration

NBA coaches make substituion based on gut feelings not data. We built an AI tool that predicts both injury and win probability so coaches can make smarter data driven decisions

What it does

Smart substitution advisor analyzes workload using machine learning to predict injury risk and win probability. Coaches select a player-- see the injury/win metric--get coaching recommendations(rest, play or rotate). Features include injury gauge, win probability charts and playing time recommendations

How we built it

Generated synthetic NBA data Engineered 15+ features Trained two ML models: Random forest(injury, ROCo-AUC 0.78) and Gradient Boosting(Wins ROC-AUC 0.82) Built interactive streamlit dashboard with plotyly charts Deployed to Github

Challenges we ran into

NBA API unreliable - used synthetic data instead Class imbalance(rare injuries) - used ROC-AUC instead of Accuracy Complex UI - simplified to step by step wizard

Accomplishments that we're proud of

Complete ML pipeline (data- train- deploy) Statistically sound models practical tool coaches can use Interactive dashboard with live predictions Professional UI with basketball background

What we learned

Domain knowledge matters more than fancy algorithmms Class imbalances require special evaluation Feature engineering is bigger than model complexity Deployment is half the work Interactive design improves usability