About the project
Racketlon match prediction, built from real data.
RateMyRacket is based on a full machine learning pipeline that predicts Racketlon matches across table tennis, badminton, squash, and tennis. It combines player ratings, recent form, and historical performance to estimate both match scores and winners.
What the model actually predicts
Instead of just predicting a winner, the model predicts the score difference in each sport, then combines them into a full match result. This makes predictions more realistic and more informative than a simple win/loss guess.
Model performance
How accurate is it?
73.4%
Winner accuracy
The model correctly predicts the winner about 3 out of 4 times.
±14 pts
Match error (MAE)
Predictions are off by about 14 total points on average.
For example, if a match ends 80–70, the model might predict something like 75–68. It captures the overall margin well, even if it’s not exact down to every point.
Interpreting the results
These ratings and predictions are data-driven estimates, not official rankings. Their accuracy depends on the amount of available match data, how recently a player has competed, and the level of opponents they typically face.
Players with fewer matches or highly uneven competition histories may have less stable ratings, and predictions for those matchups may be less reliable.
Sport-level accuracy
Per-sport predictions
TT
±6 pts
Table Tennis
BD
±6 pts
Badminton
SQ
±7 pts
Squash
TN
±5 pts
Tennis
Each sport is predicted within about 5–7 points on average. These errors accumulate across the four sports, leading to the overall match error.
Research paper
Full methodology & results
Key idea
Good features matter more than complex models.
The biggest improvement comes from how player performance is represented — ratings, recent form, and matchup history — not from using more complex machine learning models.
About the author
Built by Zain Magdon-Ismail as a Racketlon analytics and prediction project.