Ellipse reversed header logo
Insight

Introducing Oval’s Player Clustering Tool

Article title and image of player clustering.

Data-powered player description

What is the best way to describe a player? In terms of quality, you might go with ‘good’ or ‘bad’. In terms of physical attributes, popular options include ‘rapid’ or ‘powerful’ – often after the word ‘deceptive’.

“In rugby, there has been no way to objectively identify an individual’s playing style, to see how players’ roles change over time or in different teams.”

Skills are even more difficult to describe: ‘elusive’, ‘barnstorming’, ‘tactically astute’. This collection of adjectives is hardly scientific, with the meaning of each becoming more blurred as it is attributed to more players. Even the most capable coaches and most knowledgeable fans categorise players based on opinion, often guided by headline-grabbing breaks, tries, and turnovers. Does Simon Zebo’s heel flick mean he is a ‘flair player’ ten years on? Is Tom Curry really an ‘out-and-out’ seven? In rugby, there has been no way to objectively identify an individual’s playing style, to see how players’ roles change over time or in different teams.

To solve this problem, the latest in Oval’s suite of analytics products is its ‘Player Clustering’ tool. Not focused on one famous performance or looking to shoehorn a player into a particular role – how many ‘next Jonah Lomus’ have we had now? – this tool is powered by a dataset including every action by a player in a particular season and team. The Clustering tool can describe a player’s style more accurately than an adjective or the number on their jersey, and can objectively identify a player’s role in a team at any given point in their career.

Take Garry Ringrose for example. On Sunday – in possibly the toughest pool fixture in Champions Cup history – Ringrose will face Jonathan Danty in the midfield. To the naked eye, this seems like a contest of contrasts: guile versus power. However, in their performances for their clubs this season, both Ringrose and Danty are described as ‘Midfield First Receivers’ by the Clustering tool, an unsupervised machine learning model.

Oval Data Science CENTRES FINAL

Compared to other centres in attack, players such as Ringrose and Danty make a higher percentage of their carries after one pass. Typically, Midfield First Receivers (MFRs) are bruising ball carriers, given early opportunities to make dents in opposition defences with their physicality. We can see this in the percentage of their carry metres which are made in contact, which is comparatively high for MFRs. Danty is the archetype for this cluster, a 99.996% match. Since returning for La Rochelle in the TOP 14, 43% of his total metreage has been in contact, compared to 27% for centres in other attacking clusters; he works hard for extra yards in the carry.

Ringrose’s reputation isn’t as a ‘basher’, instead renowned for attacking space with more subtle methods than a Danty-esque midfielder. This is because Ringrose has occupied a variety of roles for Leinster and Ireland. For Leinster last season, he was a ‘Wide Attacker’: a much greater proportion of his carries were on the edges of the pitch, and he tended to offload more. However, this season – perhaps in the absence of Johnny Sexton – he has been utilised closer to the breakdown, taking advantage of his passing ability. Ringrose’s similarity to Danty has increased – he is more regularly at first receiver, more of his metres are made in the tackle – but he retains the profile of a more distributive centre, opting to pass more often than his counterpart.

The Clustering tool is just as perceptive looking at players’ decision making in defence as attack; it also doesn’t discriminate against less glamourous positions than centre. A seismic head-to-head this weekend is between the gargantuan duo of Uini Atonio and Andrew Porter. Both figureheads of their team’s physical effort, their defensive clusters give an indication of the ethe of La Rochelle and Leinster without the ball, and where these teams will try to target one another.

When speculating on the outcome of a battle between a loosehead and tighthead, the scrum is always front and centre of the discussion. Atonio’s cluster indicates his personal focus on the set-piece: he is described as a ‘Defensive Scrummaging Tackler’ (DST). Characteristic of props in this cluster, a high proportion of his defensive involvements are scrums. He may not be involved as often in the loose as other props, but tackles willingly and conserves energy for gladiatorial contests at scrum-time. Obviously scrummaging is a critical part of the game for all props, but it is a pronounced priority of players described as DSTs: their scrummaging accounts for an average of 5% more of their defensive involvements than players in the ‘Defensive Rucker’ or ‘Midfield Lowerbody Tackler’ categories.

“A seismic head-to-head this weekend is between the gargantuan duo of Uini Atonio and Andrew Porter.”

Oval Data Science HOOKERS FINAL

In contrast, Andrew Porter has an extremely high work rate for a prop. Of all the props to play for a URC club last season, Porter averaged the fourth-most defensive actions per 80 minutes; he also played the most minutes of the top four. Average defensive actions is the only performance-related metric used in defensive clustering, with the rest of the factors taken into consideration focused on decisionmaking. So what decisions is Porter making which are getting him involved 55 times a match? The main driver of Porter’s classification as a ‘Defensive Rucker’ is his eagerness to compete at the breakdown. After 22% of the tackles Porter made for Leinster last season, he was also the first player to enter the ensuing breakdown – tireless work.

What decisions is Porter making which are getting him involved 55 times a match?

Often head-to-heads are described as a ‘game within a game’. In lieu of their different clusters, it will be fascinating to see how these contrasting playing styles trade off. Porter’s athleticism and near-constant involvement are impressive, but will this prove a disadvantage when preparing to scrummage the giant Atonio? Equally exciting will be to see whether Atonio will be allowed to focus on set-piece, or whether the relentless Leinster attack will drag him into a contest ill-suited to his skillset.

This could become the ultimate selection and recruitment tool

The Clustering tool produces incredibly interesting player style insights, another example of Oval pushing forward the frontier of analytics in rugby. However, in combination with existing products – notably the player value model – this could become the ultimate selection and recruitment tool. How do you identify players capable of playing multiple positions, such as Beauden Barrett or Marcus Smith? How does the role of a ‘Bomb Squad’ player differ from that of a starting Springbok forward? The power of Oval analytics, in combination, allows you to pose these questions and receive simple, data-driven answers: you can identify the style of these players in context and assess their value in these roles.

In contrast, Andrew Porter has an extremely high work rate for a prop. Of all the props to play for a URC club last season, Porter averaged the fourth-most defensive actions per 80 minutes; he also played the most minutes of the top four. Average defensive actions is the only performance-related metric used in defensive clustering, with the rest of the factors taken into consideration focused on decisionmaking. So what decisions is Porter making which are getting him involved 55 times a match? The main driver of Porter’s classification as a ‘Defensive Rucker’ is his eagerness to compete at the breakdown. After 22% of the tackles Porter made for Leinster last season, he was also the first player to enter the ensuing breakdown – tireless work.

To find out more on how Oval is Powering Rugby, visit ovalinsights.com.

Share the Post: