Author: Qijie Shen et al
Journal: ScienceDirect (2026)

---

This study analysed match data from the 2023 FIFA Women’s World Cup to answer a simple but powerful question:

Can we describe how players actually behave on the pitch by combining physical output and technical actions — instead of relying on positions or single metrics?

To do this, the researchers merged:

• Physical tracking data (running volumes, intensities, movement patterns)
• Technical offensive event data (actions with the ball, attacking involvement)

They then used K-means clustering (an unsupervised machine-learning method) to group similar match performances together.

---

What They Found

From 1,599 player-match observations across 539 players, the model identified 18 distinct behavioural profiles. These profiles were not “positions” like winger or striker, but ways of playing such as:

• High-intensity attacking profiles
• Aerial target profiles
• Technically involved but physically economical profiles
• Impact substitute–type profiles

These clusters represent real patterns of match behaviour that emerge when you combine physical and technical data.

---

The Crucial Finding Coaches Must Understand

Only 184 of the 539 players stayed in the same cluster across multiple matches.

The majority of players appeared in two or more clusters across their matches.

At first glance, this might look like a weakness of clustering.

It is actually one of the most important findings of the study.

Why?

Because it shows something coaches already know intuitively:

Football behaviour is highly context-dependent.

A player does not perform the same way every match because:

• Tactics change
• Opponents change
• Match state changes
• Role instructions change

The clustering did not “fail” to label players.
It revealed that many players are behaviourally flexible.

The researchers describe this as:

• Specialised profiles → players with very stable match behaviour (184 players)
• Hybrid profiles → players whose behaviour shifts depending on context (the majority)

This is not noise.
This is the football story appearing in the data.

---

What This Means for Coaches

1. Stop Thinking in Positions — Start Thinking in Behaviours

These clusters describe how players actually perform, not where they are named on the lineup sheet.

Two wingers may belong to completely different behavioural clusters.

2. Identify Specialists vs Tactical Chameleons

You can distinguish:

• Players who are extremely consistent in their match behaviour
• Players who adapt their behaviour depending on tactical context

Both are valuable — for different reasons.

3. Training Design

Instead of designing drills by position, you can design them by behavioural demands:

• High-intensity attacking behaviour
• Aerial duel / target play behaviour
• Technically dominant but lower running output behaviour

4. Scouting Your Own Players

You can track:

• When a player’s behaviour shifts across matches
• Whether that shift was intentional (tactical instruction) or unintended
• Which players have natural flexibility

5. Scouting Opponents

This is where it becomes very powerful.

Rather than scouting opponents by formation or position, you can identify:

• Which behavioural profiles their key players typically show
• Which profiles appear when they are winning vs losing
• Which players change cluster depending on match context

You start scouting patterns of behaviour, not names on a team sheet.

---

Why Clustering Is Still Valuable (Even if Players Move Between Clusters)

Clustering is not meant to give a permanent label like:

“This player is Type 7.”

It shows:

“These are the types of match behaviours that exist — and when players display them.”

That is far more useful for coaching and scouting.

Because football is dynamic.

---

The Big Idea of the Research

This paper shows that when you combine physical and technical data:

• You can objectively describe real match behaviours
• You can see which players are behaviourally stable
• You can see which players are behaviourally flexible
• You can move analysis away from positions toward functional roles
• You can scout players and opponents based on how they play, not where they play

And importantly:

The variability in clusters is not a problem.
It is evidence that the model is capturing the true complexity of football behaviour.

Note: This summary was generated with the assistance of ChatGPT based on the original paper, with the aim of translating the research into practical insights for coaches and practitioners.