demo_nextmatch
TRANSCRIPT
• Identify variables that matter in terms of winning • Offer actionable insights for match strategy plan
Problem: Use data to impact soccer match outcome
Solution: NextMatch
“Many organizations spend considerable sums on collecting data, but lack the capability to … analyze the data using machine learning.”– Tony Strudwick, Head of Athletic Development, Manchester United FC
“(By doing simple correlation, one finds no single stats matters for winning), the only stats that matters is winning”– Jo Clubb, Lead Sport Scientist, Chelsea FC
Scraped 3 years of match stats for each team in the Spanish Soccer League from whoscored.com
Data and Algorithm
Outcome of Team A’s match
Features of team A
Features of team A’s opponent
1: win / 0: not win Passes, Shots, … Passes_OP, Shots_OP, …
Step Algorithm Advantage
1. Feature Selection
Random Forest Dealing with large No. of features while maintaining interpretability
2. Classification Logistic regression using features selected above
Simple; Easy interpretation; Sign of coefficients
A different set of models for each team
Model Performance
Model performs well on most teams
Training: 12/13-13/14 season, 76 matches / team; Testing: 14/15 season, 18 matches / team
Model Performance
Matches of teams that change squad a lot in a short period are less predictable
Training: 12/13-13/14 season, 76 matches / team; Testing: 14/15 season, 18 matches / team
Use Case – for the team Valencia
Match Valencia Sevilla
Round 2 3 1
Recommendations Valencia “followed”