clock menu more-arrow no yes

Filed under:

What is FC Dallas Missing? - Modern MLS Player Roles - Machine Learning Analysis Part 2

New, 15 comments

What makes a championship team in MLS? Our data is helping paint that picture.

Steve is learning.
iStockPhoto.com

Link to Part 1

Steve and I are back again, this time we are here to spread some knowledge and help better inform the fanbase about what types of players it takes to win a championship. Last week I broke down some basics as to how I classified modern MLS player roles with machine learning techniques using more than 100 individual statistics. We did not get very deep into any conclusions from the classification, but this week we will dive into two important actionable insights:

  1. What makes up a championship team?
  2. What FC Dallas is missing from their current team?

In this article I will refer to player groups/roles/positions by giving them a name rather than using the assigned number for the cluster. This will (hopefully) help to better describe the role and hopefully make it easier to keep track of the differences. Teams are full of players fulfilling a variety of roles and the ones I discuss in this article are no different. Within each team, there are players who reside in the same role groups which helps us infer a few things, including style of play. Before we get into looking at what FC Dallas does and does not have, and what other MLS teams are best set up to secure a championship; let’s explore what forms the foundations of the previous five MLS championship teams.

Championship DNA - What’s in a title winning team?

After running the classifier, we can see the roles that players are assigned for each season they play, as well as all of the different player roles that, when combined together, make up each team. Our dataset only contains data for the past five seasons, so it is not intensively comprehensive, but there are several patterns that appear within the data. Of the previous five MLS Cup champions each team contained no less than six of the following:

  • Attacking creator - Landon Donovan (LAG), Diego Valeri (POR), Victor Vazquez (TFC), Hector Villalba (ATL), Miguel Almiron (ATL)
  • Goal-scoring machine - Robbie Keane (LAG), Clint Dempsey (SEA), Sebastian Giovinco (TFC), Josef Martinez (ATL)
  • Deep-lying possession - Juninho (LAG), Osvaldo Alonso (SEA), Michael Bradley (TFC), Leandro Gonzalez Pirez (ATL)
  • Windshield wiper - Omar Gonzalez (LAG), Nat Borchers (POR), Cristian Roldan (SEA), Drew Moor (TFC)
  • Physical attacking specimen - Gyasi Zardes (LAG), Fanendo Adi (POR), Jordan Morris (SEA), Jozy Altidore (TFC)
  • Versatile attacking midfielder - Stefan Ishizaki (LAG), Will Johnson (POR), Andreas Ivanschitz (SEA), Ezequiel Barco (ATL)
  • Wingback - Dan Gargan (LAG), Jorge Villafana (POR), Joevin Jones (SEA), Justin Morrow (TFC), Chris McCann (ATL)

While we are on the subject of roles, I think it is important to remember that each of the roles consists of forty (40) traits that are individually made up of all of the 100+ different player statistics (weighed within each trait) that we collected for the previous five years of soccer. Different teams tend to play quite differently but are all rather similar in that they each have these important roles which contribute in all phases of play. Just below is an example of how the clustering appears (color coded) for a selected three of our many statistics. These three statistics had a strong impact on grouping, thus I wanted to show it to all of you so that you can start to see some of the trends within the data.

Many of you are likely thinking that these roles are fairly obvious, and that you would assume most high-quality teams would have players occupying these roles. Perhaps this is where the idea behind naming the roles hits a big of a snag. While the role “names” are understandable, perhaps they are a bit simplistic. Looking at the players listed, you may also notice that the groups do not necessarily contain players who play the same traditional position. That is an important insight for us as we continue this analysis! Christian Roldan and Omar Gonzalez play different “positions” but their statistical output, distribution, and patterns are all quite similar and allow us to classify them together! The same situation occurs when looking at Michael Bradley and Leandro Gonzalez Pirez, different positions but similar statistical output.

Now that we have a better grasp of some of the “molecules” which make up our “championship DNA”, we should explore some applications of this type of data. Knowing what roles are replicated across each of the championship teams, we can now dig into which of these roles our beloved FC Dallas is missing.

What is FC Dallas missing?... and who should FC Dallas target to fill the holes?

FC Dallas has many talented young players, several experienced veterans, and a mix of role players that form the basis for their own unique run at a championship. I appreciate the mixture and variety of talent, but the club is missing a few key player-roles that championship teams tend to have.

When we are looking at current rosters in Major League Soccer and analyzing what each team does and does not have, we have to take a few things into account - namely that we have very little data for 2019 and that there are many changes from 2018 to 2019. We can interpolate some data (fill in gaps between known datapoints) and we can extrapolate other data (extend our dataset beyond known datapoints), we can also examine common patterns and outliers; and using some mathematical principles we are able to predict what 2019 will probably look like for the players in MLS, within a certain range of likelihood.

For the purposes of this article I am going to stick to the seven player roles listed above because a more full explanation would be more fitting in a white paper with deeper analysis. Because I am interested, I want to show what FC Dallas does have first, and then dig into what they are missing.

  • Wingback - Two words here: Reggie. Cannon. Guy is a beast and fits this role perfectly.
  • Attacking creator - Paxton Pomykal fits here, but only just. Depending on how we predict his entire 2019 season he sits just between this role and a deep-lying playmaker role. Luchi’s triple pivot (see Jason Poon’s appearance on the Dallas Soccer Show podcast for more info) puts Paxton into a unique spot to serve more than one purpose and dynamically show up all over the field.
  • Windshield wiper - Here we find Matt Hedges, who (ugly dance with Cory Burke aside) has performed quite well and looks rather good in the more aggressive role. Also Reto Ziegler falls in here in 68% of the 2019 projections, so we should consider him as well.
  • Deep-lying playmaker - Carlos Gruezo is forming into quite the valuable member of a trio that pivots in and out of the deepest, middle, and highest points in the FC Dallas midfield triangle. His role tends to be deeper than any of the others and his statistical output aligns with this role.

Looking over the above list, FC Dallas does have some important pieces, which is promising but they are still lacking some key “molecules.” While the team is missing these roles, there are some players who could, by season’s end or further into the future, evolve into these absent roles. Here I am going to list some players that I think would be good fits with FC Dallas in these key areas, as well as any currently rostered players that could also fill in those gaps.

  • Goal-scoring machine - These guys are incredibly hard to find and frankly, acquiring one from within the league is not going to be easy. The players within the league who fit this role are: Wayne Rooney, Zlatan Ibrahimovic, Josef Martinez. As far as I can see FC Dallas does not have anyone on the roster who could turn into this type of player. North Texas SC on the other hand...maybe?
  • Physical attacking specimen - Thankfully, this type of player is easier to find, and when coupled with a creative type midfielder, goals do not seem to be an issue. Currently rostered players that could turn into this are: Badji, Cobra, and possibly Cristian Colman (yes, he is still here). If we are looking within the league and outside of the club then we have plenty of options! Nemanja Nikolic (Chicago), Daniel Royer (NYRB), Mauro Manotas (Houston), Christian Ramirez (LAFC), Dom Dwyer (Orlando), Kei Kamara (VWFC), oh and look for Cyle Larin should he return to MLS. I do not think he would be the first choice for the FC Dallas front office, but he remains an interesting option.
  • Versatile attacking midfielder - What is interesting here is that these players are not dynamos and do not appear crucial in any statistical manner for the championship team but they are vital to the opportunities that are created for other players on the team and thus, invaluable. Players in the league (tons exist so let’s just examine a few interesting ones): Kellyn Acosta (COL- guessing this is a no), Pedro Santos (CLB), Sean Davis (NYRB), and Felipe (VWFC). From within - players like Jesus Ferreira (yes, I know he is traditionally a striker, but his stats and patterns of play trend toward this type of midfield or wide role), Ryan Hollingshead - who is a great utility player, and young Mr. Thomas Roberts could all develop into this role without much straying from the current pattern. Roberts also has potential to develop into other roles but limited data makes his projection much more difficult.

From this small bit of analysis we can see that there are some interesting patterns and roles filled by players on a variety of teams. There are also plenty of players who could fit into these roles for FC Dallas both from within MLS, where our current dataset exists, and outside, where there is plenty of data available for us to include at a later date. Just below here I’ve put the top 5 “traits” for each of the roles listed today.

Which current teams are most similar recent championship teams?

I will end with this and give you guys the top five teams that are most similar in makeup to the previous champions of MLS Cup in roster makeup. I determined similarity using several different comparison methods and weighted them to come up with a custom similarity index. In order:

  1. Philadelphia Union
  2. Toronto FC
  3. Columbus Crew
  4. Real Salt Lake
  5. DC United

FC Dallas came in tenth on the list, just ahead of Orlando City and just behind Sporting Kansas City for those that are curious. If you have any questions, please ask away and be sure to comment with any feedback or opinions you have as well! Next week will bring part 3 of this series and some further analysis into predictive analytics - using this data to project player development and growth, as well as player evolution with age, team, minutes, etc. I hope to see some comments and drive some discussion here so comment and share this around!