Very interesting. The top 16 teams with better prob to make the second phase were almost the same that actually advanced. The only issue I see is using the FIFA ranking as it is not a very good way of actually ranking the teams. Great job.

From what I can see most of the analysis so far is taken from "static" data, ranking, host nation advantage, size of midfielder etc.. the cup is an evolving competition where highs from a previous game can soon turn to lows e.g. germany/sweden high followed by germany/south korea low or where injuries incurred in previous games can affect whether key players can play in subsequent knockout matches e.g. Uruguay's Cavani, and therefore negatively influence the result, or whether yellow cards given in previous matches can have a similar affect on subsequent matches. I can imagine that if on-going "live" data was fed into the computation we would end up with an evermore plausible winner :-) As seen by Apps like "Tor alarm" (german for goal alarm) live data such as: team makeup, substitutions, when the game starts/ends, yellow cards, goals and by whom is available online.

Germany lost, who is the next champion?

Yes, we can do a whole set of comparisons for the countries playing the world cup.

About the players data, very nice data, but anyway it is always different how they play in their national selections than in their regular teams. And no match is decided until it is played!

Anyway I wonder if we could also classify the teams by how well they fit into that model of player features by position. Is a team with all its players closer to the avg of that plot a better team? That kind of questions could be interesting to compute.

This is great! @Francisco thank you for sharing! I wonder how your predictions compare to this recent article:

Prediction of the FIFA World Cup 2018 – A random forest approach with an emphasis on estimated team ability parameters https://arxiv.org/abs/1806.03208

which was also covered in MIT Technology Review blog: Machine learning predicts World Cup winner. There basic result is the following tree of wins:

During the past World Cup a close analysis was done by @Etienne Bernard with very neat machine learning techniques:

• Predicting Who Will Win the World Cup with Wolfram Language

• World Cup Follow-Up: Update of Winning Probabilities and Betting Results

The similar visualization was: