Using Machine Learning to replace Ben Chilwell at Leicester

Updated: Sep 3, 2020

With the recent departure of Ben Chilwell from Leicester City to Chelsea for £50million, Touchline Analytics wanted to explore potential replacements, and how data and Machine Learning could be used to expose some players that might otherwise be overlooked. By using Principle Component Analysis, we can assess players across a multitude of metrics, to identify players who are similar in playing style. We limit the analysis below to the English Premier League, English Championship, Scottish Premiership, Dutch Eredivisie.

Firstly, let’s take a look at Ben Chilwell across a number of key metrics, and see how he compares to the player he is going to replace at Chelsea, Marcos Alonso.

Ben Chilwell is strong across many of the key attacking metrics involved in being a full back, in particular, travelling with the ball. However, in the same vein as Marcos Alonso, Chilwell’s defensive game isn’t strong when compared to other left backs.

We can start by taking a look at how Chilwell compares to other left backs across a handful of European leagues, using 8 key metrics that encompass the attacking, defensive and possession aspects of a full backs game. For each of these metrics we can assign players into deciles, ranking players into 10 buckets, assigning the top 10% of players a score of 10, the next 10% of players a score of 9 and so on.

Looking deeper into Chilwell we can see that across the attacking and possession based aspects of the game, Chilwell excels. Whilst, looking into the defensive side of the full back role (aerial duels aside) Chilwell is not as strong.

Comparing this to Marcos Alonso, the player who Chilwell will replace at Chelsea, we see a similar story. Alonso is particularly strong across possession and attacking based metrics, in fact, outperforming Chilwell in many. The defensive side of Alonso's game is not an area where Alonso excels, although he also doesn't struggle, fitting into the higher side of average.

Looking across previous seasons we see that whilst Chilwell is still improving his game (he's still developing at the age of 23), Alonso (29) has hit his peak and is no longer improving across these key metrics. Chelsea might see the opportunity to replace Alonso with a player who is similarly effective in attacking, but with the ability to further improve the defensive side of the game. It also gives Lampard the chance to play 4 at the back with Chilwell more comfortable in that position.

Leicester’s style of play relies on both full backs attacking so when looking for a

replacement it will be important to identify players similar to Chilwell, or look at players who over index in attacking metrics.

By using Principle component analysis (PCA) we can distil down information from many metrics (such as the 8 above) into just two, allowing us to plot on a simple x-y plane and segment players into groups with similar attributes.

On their own the two principle components don’t mean anything, but players with similar x-y values have similar playing styles, allowing us to visually split out players into different styles, purely based on the data. When we then look at these players in our PCA analysis (below chart), Marcos Alonso and Ben Chilwell are close in terms of playing styles.

We see 4 different clusters on the below chart and they vaguely relate to the following segments:

Red: Over index on defensive metrics Green: Over index on attacking metrics

Orange: Well rounded Blue: Don't excel across the metrics selected

Leveraging the output from the PCA we can take a deep dive look into the players that are closest to Chilwell, these are the players who are most similar to him across the metrics we have chosen "Like for like replacements"

All of these players show great promise as attacking left backs, with some of them also having more favourable defensive statistics as well. The closest players to Chilwell and Alonso, in the same cluster are:

Ryan Manning (QPR, 24)

Joe Bryan (Fulham, 26)

Max Clarke (Vitesse, 24)

Nicolas Taglifico (Ajax, 27)

Along with the following who we won’t look further into for various reasons (already performing in the premier league or recently moved teams in this window).

Antonee Robinson (Fulham, 23)

Oleksandr Zinchenko (Manchester City, 23)

Luke Shaw (Manchester City, 25)

Lucas Digne (Everton, 27)

We can also use the output of the PCA to take a look into other players that are interesting. Looking at players that are outliers within the same cluster as players of interest.

We see one player, Borna Barišić (Rangers, 27), within the green cluster (over index in attacking metrics) that is an outlier, far away from all other players. We can also take a look into this player to understand more how likely they would be to fit into the Leicester style of play.

Investigating further into the players the PCA identified, we are going to consider two in more detail below

Each bar in the charts below represents a player with all players ordered from best to worst (left to right) within each metric. We then highlight the two players we want to compare to see how they stack up against each other, and the rest of the players we have input into our PCA.

We could compare these metrics on a radar chart, superimposing players on top of each other based on their decile rank. However, the below allows us to also compare the variance within the key metrics, for example highlighting just how strong Chilwell is at carrying the ball when stacked up against the other left backs being considered.

Ryan Manning || QPR || 24yo

Comparing Manning and Chilwell we see that they show fairly similar profiles across these three key areas. Where Chilwell is strong at carrying the ball, Manning's game centres around using the ball to create chances for others, over indexing against Chilwell for both the number of crosses per game, and because of this, the number of shots assisted per 90. Defensively, Chilwell is stronger in the air, whilst Manning excels at ground duels.

Joe Bryan || Fulham || 26yo

Joe Bryan, whilst in the same cluster as Chilwell, was closer to the 'strong defenders' cluster than Chilwell and this shows within the statistics below. Bryans game shows similar distribution of crosses and chances created, however the possession statistics show an interesting story. Despite having a lot fewer successful passes per 90, Bryan uses the ball better when passing into and when within the final third, with a higher proportion of these passes into the final third successful. Defensively, Bryan is stronger in overall duel success %, despite not being as strong aerially.

We have used the data to create a shortlist of players, turning 100s of potential candidates into a manageable list. These players can then be scouted in detail allowing the clubs back room staff to focus in on players who are most likely to be good replacement.

In the above, we're comparing players across different leagues, and from teams with different playing styles. By choosing players from teams who play in a similar style and from leagues who we know players have successfully made the step up from, we eliminate some of these potential issues.

143 views0 comments

Recent Posts

See All

Applied Tableau Training for Football - An Introduction Just Launched: The first ever course from a Football specific Tableau Partner. We've leveraged the knowledge of Tom Goodall and his years of experi