🧊 EndGame Theory: Optimal Strategies on Ice

💡 Inspiration

Curling feels like chess on ice, yet strategies often rely on intuition.
We wanted to find out when and why certain decisions — like Power Plays or blanking — are truly optimal.
So, we combined game theory and data analytics to model curling as a strategic game between two rational players.

🧠 What We Learned

How to quantify the value of the hammer and model win probabilities.
When blanking an end actually improves winning chances.
How different nations show unique shot patterns and risk styles.
How to use Markov Decision Processes (MDP) to simulate tactical outcomes.

🏗️ How We Built It

Merged Games.csv, Ends.csv, and Stones.csv
Engineered features: hammer flag, score differential, shot success, PowerPlay indicator
Used Python (pandas, PySpark, scikit-learn) for modeling and Plotly for visualization
Modeled shot transitions with an MDP framework

🚧 Challenges

Handling complex stone coordinate data
Dealing with small samples for rare scenarios (Power Plays, final ends)
Translating game theory payoffs into real curling metrics
Making results both mathematical and coach-friendly

🧩 Results

Blanking the 7th end often increases winning odds
Optimal Power Play when trailing after 5th end
Country insights:
- 🇨🇦 Canada → Efficient, conservative
- 🇸🇪 Sweden → Defensive, precise
- 🇰🇷 Korea → Balanced and adaptive
- 🇳🇴 Norway → Aggressive, high-risk

🏁 Takeaway

We built a data-driven and game-theoretic model that transforms curling analysis from descriptive to prescriptive.
This framework can help predict the optimal shot choice for any in-game situation — the future of smart curling analytics.