Weird Science - chopping up teams with data

All data in this blog is provided by Football Whispers.

Imagine you could take a successful football team, break it down into the individual players, find the nearest match for that player and then put it back together.

Would you end up with Kelly Le Brock from Weird Science or, more likely Frankenstein's monster?

Kelly Le Brock from Weird Science, google it if you are young.


To give it a go I have picked Manchester City's normal team, which in my mind is:

Ederson
Walker
Stones
Otamendi
Mendy
Fernandinho
B Silva
D Silva
Sterling
Aguero
Sane

I'm going to use data from Football Whispers to find the nearest equivalent player (excluding team mates) to make up this team. As goalkeepers can't be easily compared we'll ignore them and allow Ederson to play for both.

This will be the Kelly Le Brock XI made up of the players with the most similar stylistic profiles to that Manchester City line up above.

I have picked the player with the nearest stylistic match, it doesn't mean they are as good as the other player (although a ranking for that is also provided).

And the resulting XI is:

Ederson
Elseid Hysaj
Thiago Silva
Niklas Sule
David Alaba
Javi Garcia
Javier Pastore
Mesut Ozil
Anthony Martial
Luis Suarez
Paulo Dybala

As a quick sanity check of the concept of profiling players for similar statistical markers I think it passes. There aren't any of those players I look at and think "they are completely different types of players". Ironically with Fernandinho needing replacing soon the closest match was an ex-Manchester City midfielder in Javi Garcia.

All in all I'd say that works as a concept.

Just so poor Kevin De Bruyne didn't feel left out I did a special one for him and got Eriksen as the nearest match. Again fair enough, but it also suggested Kevin Stoger, not someone I'd heard of.

So on to the more fun version.



My Frankenstein's Monster XI is simply made up of the first player on the list of stylistic matches who I either hadn't heard of, or was surprised they appeared. I also restricted it by age to a maximum of 23.

Ederson (because we can't do this for goalkeepers)
Kevin Diks - Fiorentina right back from Feyernoord
Pantelis Hatzidiakos - AZ Greek-Dutch defender
Marlon Santos - ex Barca now with Sassuolo
Thomas Ouwejean - AZ attacking left back
Victorien Angban - Metz, Chelsea never ending loanee
Daniel Crowley - Willem II, thriving in Holland
Simon Gustafson - FC Utrecht, Swedish attacking midfielder
Ousamma Idrissi - AZ (again) all round tricky attacking winger
Musa Barrow - Atalanta, very high on creativity in limited minutes which might be why he appears
Myziane Maolida - Nice, tricky wide attacker

The £ shop De Bruyne is Andre Horta, now of Los Angeles FC!

A lot of Dutch based, or raised, players appear on this list, *strokes chin knowingly* perhaps reflecting the similarities between the Guardiola style of play and the Dutch development practice of Rinus Michels.

Whatever it might be there is obviously a stylistic match between Manchester City and the way teams play in the Eredivisie with half this team based there. Probably true of Marlon Santos too who spent some time with Barcelona picking up that style of possession play.

Angban seems to be a Chelsea loanee I've never heard of, he seems to be doing OK in the French second tier, winning the ball lots and passing reasonably.

Musa Barrow is an interesting player too as he, along with Ousamma Idrissi, come up as very close matches to Messi in style (obviously with WAY lower performance scores).

Maolida looks great from limited minutes with similar ball carrying and cutting in on goal that I associate with Sane.

I stress again there were other players U23 who were closer matches to some of the above choice - for example Frenkie De Jong was a closer stylistic match and had a much higher performance score than Angban so hipsters can breathe again.

The Frankenstein XI is more a "can this system pick up relative unknowns" and it can.

I shall continue my experiments and see what other unknowns I can dig out using the stylistic matching data.

Comments

Popular posts from this blog

Wyscout review and poking around the French third tier

Scouting report Dan Ndoye - Lausanne Sport

Data Analytics conference - Daniel Krueger report