AthleteLoadFootball performance research View source

Workload. Recovery. Context.

The training
picture.

Heavy legs. A short night. Another hard session.

See how the pieces add up. Adjust an athlete’s training and recovery to explore the signals behind injury-risk research.

SoccerMon study2020–2021 / Women’s elite football
01 / Scenario lab

Change the load.
Read the response.

Illustrative simulation
Changes update instantly

Start with a scenario

Your inputs

Recovered scenario
Training load
1,600 sRPE
2004,500 sRPE
1,700 sRPE
4004,000 sRPE
Acute : chronic ratio
0.94 / Balanced
01.53.0+
Recovery & wellness

Self-reported ratings, on a 1–10 scale unless noted.

2 / 10
2 / 10
8.5 hrs
8 / 10
8 / 10
2 / 10

The response

Formula-based illustration
7-day risk estimateLow
0.5%

Illustrative probability, not a clinical prediction

Next-day readinessHigh
9.0/ 10

Based on the recovery inputs at left

Recovery in balance

Explore a different scenario to see how the response changes.

What’s moving the estimate?

Contribution to simulated log-odds. Longer bars have more influence.

− Lower estimate+ Higher estimate
View the 3D strain illustration Conceptual body map

Where load is illustrated

This simplified body map responds to workload, soreness and fatigue. It does not localize or predict an injury.

Illustrated focus
Lower body
Workload state
Balanced

Open the illustration to load the 3D view.

Drag to rotate · Scroll to zoom

About this lab: the sliders use illustrative browser formulas, not the trained XGBoost model or TreeSHAP. The published model results are reported separately below.

02 / Athlete history

A session is only part of the story.

Two seasons of training context.
Real SoccerMon athlete records.

Loading athlete history…

Daily loadAcute loadChronic loadACWR (right axis)Injury report
03 / The research

Show the evidence.

Read the methodology ↗

A rare event.
A difficult prediction.

These results come from a chronological test split. Precision-recall AUC is the primary comparison because injury windows are uncommon.

The interactive lab above is an illustration. These are the actual exported research metrics.

Download model results ↓

Model comparison

Held-out test set
Model PR-AUC ROC-AUC Recall Precision Brier
Loading exported model results…

PR-AUC: higher is better. Brier score: lower is better. Highlight indicates the highest test PR-AUC.

View the pre-injury statistical signals
Marker Pre-injury median Healthy median Cliff’s delta Effect
Loading statistical signals…

Mann–Whitney U comparisons from the repository’s statistical exports. Association does not establish causation. Download full results.