Using Machine Learning To Find A Jules Koundé Alternative for Tottenham

How The Season Went for Tottenham Hotspurs

Tottenham finished the 2020/21 Premier League season in a measly 7th position, and just one point ahead of Mikel Arteta's heavily criticized  Arsenal. It was a season of many ups and downs, with an opening day defeat to Everton followed by an 11 game unbeaten run that saw Spurs tipped as one of the surprise contenders for the title.

The run included 2-0 victories over Arsenal and more importantly Manchester City, and it seemed like Mourinho's defensive tactics were back in full effect as Spurs cared little about possession or territory and only cared about the results, even if it meant Kane playing in the area of a central midfielder.

The overperformance of their attack is quite well advertised, so let's not focus on that and get to the defence. Mourinho's most preferred CB partnership was that of Eric Dier and Toby Alderweireld, who got 28 and 25 90s respectively. Apart from them, their other options were an error-prone Davinson Sanchez and youngsters Joe Rodon and Japhet Tanganga.

Now with Jose Mourinho gone and the appointment of Nuno Espirito Santo, Spurs have decided to focus on and rebuild the defence.

CB Options and The Chase for Jules Koundé

Spurs not only had a long list of managers they were linked to, but also one for center-backs. They have so far been linked to Takehiro Tomiyasu from Bologna, Joachim Anderson from Lyon (who was on loan at Fulham for the 2020/21 season), Conor Coady, and who is probably their most far-fetched target, Jules Koundé.

The 22-year-old is certainly the gold standard for all their targets and the one who has the highest potential. However, with Spurs only qualifying for the UEFA Conference League and Koundé being linked to the likes of Real Madrid (who might be about to lose Ramos AND Varane in the same window), this deal doesn't look like it will go through.

 

The Athletic are claiming that Spurs are ‘working’ on a deal to sign Tomiyasu and would likely get the signing done under £20m. Joachim Anderson seems like the most likely player to join the Tomiyasu as of now, but I had another idea - how about using a machine learning algorithm to find a player similar to Koundé!

What Kind Of A Player is Koundé?

                            

The Frenchman is a truly excellent defender and is not an old-fashioned defender in any sense. Standing at just 1.78m, which is pretty short for a center-back, Koundé still manages to win an impressive 3.6 aerial duals per 90 with a success rate of almost 64%. He is also excellent at progressing the ball through both passes and carries, and also supremely comfortable on the ball, with almost a dribble per game and completing ~70 passes per game.

It should then come as little surprise that Koundé is valued at £54m and Sevilla would certainly demand in excess of  £60m from Spurs, and as previously mentioned, with interest from the likes of Real Madrid, a move to Tottenham makes little sense. Hence, I decided to make use of a popular machine learning algorithm to suggest an alternative to the Frenchman.

K-Means Clustering And Data Preparation

The objective of K-means is simple: group similar data points together and discover underlying patterns. I decided to get the statistics of all players from FBRef (top five leagues only) and feed it to the algorithm so it can look for similar patterns as that of Koundé.

To achieve this objective, K-means looks for a fixed number (k) of clusters in a dataset. The K-means algorithm identifies k number of centroids, and then allocates every data point to the nearest cluster while keeping the centroids as small as possible. The ‘means’ in the K-means refers to averaging of the data; that is, finding the centroid.

We know that goals and assists numbers aren't an effective metric to judge center-backs, and some other statistics would be much more telling of a player's quality. Here are the stats I decided to consider while creating the model:

'PassCmp/90': Passes Completed Per 90

'PrgDistPass/90': Progressive Passing Distance Per 90

'LongCmp/90': Long Passes Completed Per 90

'ProgPass/90': Progressive Passes Per 90

'TklW/90': Tackles Won Per 90

'Interceptions/90': Interceptions Per 90

'Clearances/90': Clearances Per 90 

'PrgDistCarry/90': Progressive Distance Through Carres Per 90

'AerialsWon': Aerial Duals Won Per 90

'AerialWon%': Aerial Duals Win Percentage Per 90

After standardizing these numbers, it was time to figure out how many clusters should be used for the K-Means algorithm. Using an arbitrary number or using our own judgment is not a good practice for this, so I decided to use a common technique to solve this conundrum - the elbow method.

This is what it looks like:

Looking at this, it looks like the ideal number of clusters is 3, so that is the number of clusters I used to create my ML model.

The Results

Here is a sample of what the clusters obtained by a K-Means model look like (Koundé's cluster is 2) :

PS. Please note these are scaled values, hence there are so many negative values.

Let us now have a look at the center-backs suggested by the model -

Nayef Aguerd    Manuel Akanji    Joachim Andersen    Waldemar Anton    Loïc Bade    Alessandro Bastoni

Sven Botman    Jason Denayer    Rúben Dias    Axel Disasi    Gabriel    Jean-Kevin Duverne

Wesley Fofana    Marco Friedl    Marvin Friedrich    Mario Hermoso    Rob Holding    Roger Ibanez

Marc-Oliver Kempf    Presnel Kimpembe    Jules Koundé    Maxence Lacroix    Clément Lenglet   

Philipp Lienhart    Matthijs de Ligt    Lyanco    Gianluca Mancini    Facundo Medina    Benoît Badiashille

Stefan Posch    William Saliba    Milan Škriniar    Çağlar Söyüncü    Jonathan Tah    Edmond Tapsoba

Pau Torres    Dayot Upamecano    Adam Webster    Kurt Zouma

We can straight-away discount some of the players in the list due to them being too expensive, at rival clubs, or just unattainable due to other factors. Delving similarly into some other factors, here are three players I feel could be realistic targets from Spurs -

Maxence Lacroix (Wolfsburg)

The 21-year-old took massive strides in the Bundesliga last season and is now seen as one of the most sought-after talents from the seemingly endless center-back talent factory France produces. Valued at £25.20m by Transfermarkt, the 6'2'' defender would be a coup for Tottenham and might level up the Spurs defence.

Benoît Badiashille (AS Monaco)

Another Frenchman on the list, the 20-year-old is a fantastic center-back prospect. He was linked to Manchester United over the past 12 months but opted to stay and gain valuable experience in Ligue 1. Also valued at £25.20m, Badiashille is a left-footed center-back and would complement the right-footed CB list Tottenham already possess.

Jonathan Tah (Bayer Leverkusen)

The now 25-year-old German was once one of the biggest prospects in the Bundesliga. His development might not have been as steep as the player would have wanted, but is valued at just £19.80m by Transfermarkt, and hence the 6'3'' defender might prove to be a solid option for Tottenham. 

Conclusion

What started for me as a means to understand K-Means Clustering turned out to be a fun way of looking at transfer targets. Plenty of fun names like Adam Webster (who impressed for Brighton) and Sven Botman (probably the most in-demand center-back on the list) popped up, along with some players who I hadn't heard of like Rennes' Nayef Aguerd and RC Lens' Loïc Badé. I also deliberately chose to stay away from the likes Pau Torres, Alessandro Bastoni, etc. who have been very well explored in recent times.

This might not be the best way to evaluate players - or the correct way to evaluate players - but could be one of the reliable ones when refined further and with the evaluation done with more skill and experience. From a Tottenham perspective, with the news surrounding Harry Kane & the arrival of a new manager in Nuno, add to that an impending rebuild, it might be a strange season full of surprises.

References

Market Values - Transfermarkt.co.uk

Stats - fbref.com