Whose Data Trains Sports AI? A Review of 58 Studies Reveals a Youth Gap

Headlines about artificial intelligence predicting athletic performance or injury risk are everywhere. But it is worth asking a simple question: whose data were these models built on? A 2019 systematic review of 58 AI studies in team sports found that 76% of the athletes studied were professionals, only 10% were youth academy players, and just 3% were female. For coaches working with young athletes, the evidence base is still thin.

What the study looked at

Claudino and colleagues ran a systematic review of PubMed, Scopus and Web of Science, looking for studies that applied AI techniques or methods to team-sport athletes. They set out to answer two questions:

• Which AI approaches have been used to study sport performance and injury risk?
• Which sports are using which techniques?

In total, 58 studies met the inclusion criteria.

Key findings

1. Eleven techniques across twelve sports The review identified 11 AI techniques or methods applied in 12 team sports. The most common were artificial neural networks, decision tree classifiers, support vector machines and Markov processes, all of which reported good performance metrics. Soccer, basketball, handball and volleyball had the most applications.

2. Three quarters of the research is about performance Of the 58 studies, 43 (74%) focused on sport performance and 15 (26%) on injury risk. Within the performance studies, 88% dealt with technical and tactical analysis.

3. The athletes studied were mostly male professionals Studies that reported sample size covered 6,456 participants in total:

Level	Share
Professional	76%
Youth academy	10%
Collegiate	9%
Semi-professional / U18 national team	5%
Retired professional	2%

Ninety-seven percent were male and 3% female. Soccer (26%) and basketball (22%) players made up the largest groups.

4. Promising, but still unproven The authors conclude that AI use in team sports is growing and the outlook is promising. However, further evaluation research using prospective methods is needed to establish how well specific techniques actually predict outcomes.

How to read this in practice

The review does not say AI is useless. What it shows is that most published AI findings come from data on adult male professionals.

For coaches in youth and school sport, a few cautions follow:

• Don't assume a model built on professional data will transfer to teenagers or female athletes
• "Good performance metrics" were reported within each study; your own team may not see the same accuracy
• Most AI work focuses on technical and tactical analysis, with comparatively little on physical fitness or day-to-day readiness

The review has its own limits. It covered team sports only, so it is unclear whether the findings apply to individual sports, and sample sizes varied widely between studies. The authors attribute this partly to the high share of elite athletes, as large samples of professionals are hard to obtain in some sports.

Using ATHCON

Before AI can help with younger athletes, coaches need consistent data collected the same way over time — and that groundwork can start today.

In ATHCON, fitness test results are logged with the change from the previous test, team rankings and A–S grades. Athletes check in daily from their phones, and each athlete's readiness is compared against their own 14-day baseline and shown in four levels. Building up testing and daily records in a consistent format gives you material for decisions that fit your own team.

Takeaways

• Of 58 AI studies in team sports, 74% focused on performance
• 76% of athletes studied were professionals; youth academy players were 10% and female athletes 3%
• Treat AI predictions with caution, and start by collecting your own team's data consistently