2017 05 07 - guanpu/soccer GitHub Wiki

Things done:

  1. Some simple weka classifier/clustering test only including player ranking data as training data( the result is not good)

For example:

  • When we chose the average of the 11 players of each team as input, it shows connection with match result but not very rigid
                       Cluster#
Attribute   Full Data         0         1         2         3
            (25979.0)  (1735.0)  (7148.0) (10793.0)  (6303.0)
=============================================================
homescore     64.8431   10.1219   65.7352   71.6876   67.1741
awayscore     64.7232     7.203   69.8858    68.724   67.8511
result            Win       Win      Lose       Win      Draw


Clustered Instances

0       1735 (  7%)
1       7148 ( 28%)
2      10793 ( 42%)
3       6303 ( 24%)
  • While for the model of including all 22 players ranking as input, it just predicate the final match result with a possibility of roughly 50%.
  1. Refine data, specially added the data from openfootball, so that every team has data of the ranking of last several seasons, which can be added into training data, as it reveals the power of team or expectation of player.