Mens Tennis Forums banner

Predicting Djokovic's Results at the French Open

3K views 27 replies 13 participants last post by  philosophicalarf 
#1 · (Edited)
Using Djokovic's previous results at the French Open (placing heavier emphasis on his results there since 2011, which is when he had that incredible season) I attempted to create a model that would predict the score of his matches at the 2016 French Open.

Here's how it did:
Round of 128: Prediction: 63 63 62 Actual: 64 61 61
Round of 64: Prediction: 63 63 61 Actual: 75 63 64
Round of 32: Prediction: 62 63 64 Actual: 62 63 63

The model correctly predicted the winner (not very impressive) and the number of sets in each match. In the first round, it was off by 2 in the game difference between Djokovic and Lu. In the second round against Darcis, it was off by 4 in the game difference. In the third round against Bedene, it was only off by 1, even correctly predicting the score of the first two sets.

Model's prediction for match against Bautista Agut: 63 76 67 62. (4 set win with a game difference of 7)

The two main factors are the opponent's rank and the round in which Djokovic plays the opponent. Using this it comes up with the predicted winner and number of sets. Using the predicted winner, number of sets, opponent's rank, and round it comes up with the game difference Djokovic will win/lose by. It then tries to come up with a score by allocating the game difference into the number of sets.

Full post: Predicting Djokovic's French Open Run
 
See less See more
#2 ·
In b4 Djokovic flattens RBA in 3 sets 6-3, 6-1, 6-4
 
  • Like
Reactions: atennisfan
#4 ·
Djokovic d Agut 6-4 6-3 6-1
 
#27 ·
Hmm I'm considering it I've been thinking of what for project I could use. Why do you use R for data analysis? I'd also need to retrieve some more data maybe even stuff like winners/UE
A lot of the things I should have done with a Python module like scikit learn as philosophicalarf said I ended up doing with R out of convenience. Yea, you have to be careful and find the factors that are statistically significant; don't wanna complicate the model unnecessarily.
 
#11 ·
Djokovic 6-2 4-6 6-4 6-1
 
#15 ·
Uncle Latso a few posts above did!

This sounds interesting will look into it when not on mobile. What bout next match op?
6-1 7-6 4-6 6-3 for Djokovic. Not feeling too confident about that :smash: . Hopefully Nole does me the favor and keeps it close.
 
#16 ·
Could totally see that score tbh
 
#19 ·
Novak wins 6-4 6-3 6-3
 
#23 ·
Do you also take h2h or previous matches into account and maybe also similar players if there are no h2h? I am currently learning Python myself and this seems like a really interesting project. ESPN developed a method for NBA as well, BPI and it predicted pretty accurately apparently
 
#25 ·
I don't take h2h into account. It's a balance of finding the right factors to maintain statistical significance and my knowledge of statistics is rudimentary at best.
It's definitely a cool project to look into. My model was made significantly easier by the fact that it's predicting just one player's score at one particular tournament. Extending this to the point where you can just plug in two random players at any tournament is quite some work I think.
Let me know if you pursue it. Maybe we can work on something together :)
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top