Since the state of the analysis is paralysis, I wanted to do my share. Following uses a very basic +/- model with few adjustments(for minimum games played/unplayed) over 3 seasons for Galatasaray. The twist is rather than use goals, it uses ELO points from clubelo.com. Not all goals are equal; so it tries to adjust for that. Other than that it has many failings:
1-Black box- and not Shapley which might be more appropriate
2-3 season of data used to get over noise
3- Starting 11 used as I do not have dynamic Elo numbers(during the game). It is doable but this was easier. So impact subs, etc. are not taken into consideration.
4- I still prefer counting models over agnostic models but if I had a good agnostic one I would try to create a composite one.
On the other hand:
1-Passes the eye/nose test since I probably watched 95% of Galatasaray games that are in this sample.
2-Just food for thought for others with more time/computing power/better statistical skills.
Basic formula: Avg. Elo Points in games player started- Avg. Elo Points player did not start
Adjustment: If player started fewer 10 games avg. Elo point for team is assumed for games he is missing and vice versa for games he didn’t start.
If you have watched any Galatasaray games recently the team has been based on Muslera and Sneijder. Good news is most horrible players are gone or never start except for Olcan.
I know this appeals mostly to Galatasaray fans but EPL ppl can give it a try to create something similar.