I have a problem where I have to predict the winner of a game. There could upto 10 players in each game and there is always one and only one winner. Each player also gets a rank in the game (the winner is obviously ranked 1 in the game).

I have a dataset with the history of hundreds of games and the winner of the game and the rank of each player in the game).

I've structured this problem as a pair-wise 1x1 comparison of players. So if there are three players A, B and C, I evaluate the winner in each pair (A-B, B-C and A-C).

I'm using classification techniques (Random Forest, SVM etc.) to predict the winner in each pair.

My problem is, how do I predict the winner of the game, given the pairwise results. I need to consider issues like:

  1. Pairwise comparison results might be inconsistent (ie. A>B, B>C but also C>A)
  2. High winning probability in all pairwise comparisons should lead to a higher probability of winning the match
  3. Winning against a "strong player" should influence the winning probability more than winning against a "weak player"

There are some ideas mentioned here and here, but I'd like to ask if there are better approaches for my problem.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.