Availability risk
Injury Prediction Model
Predicting significant absence from workload and schedule patterns.
A multi-tier model that flags availability risk per league - 7 in 10 players in its top risk tier go on to a significant absence.
What it does
The model reads rolling workload, schedule density, and match-level context to estimate the probability that a player suffers a significant absence. It is layered - player profile, schedule context, and workload patterns each contribute - so the risk estimate reflects how injuries actually accumulate rather than a single static attribute.
Used in: Injury risk assessment.
Why it’s defensible
What makes this proprietary.
A research finding most models miss
Our research showed injury patterns are league-specific: a model trained on one competition underperforms when applied to another. So we calibrate locally - the model for one league is genuinely a different model from the next - and each beats a single pooled global model. That finding, and the infrastructure to act on it, is proprietary.
Layered features, measurable lift
The architecture is tiered - base profile, schedule context, and full workload - and each tier adds measurable accuracy. The top tier reaches roughly twice the precision of the underlying base injury rate, which is the number that matters when you are deciding whether to commit to a fragile player.
Built to be acted on
The output is a ranked risk tier, not an undifferentiated probability, so the highest-risk players surface cleanly. That ranking is calibrated against real injury data across multiple seasons.
How we validate it
Risk tiers are validated against observable absence data across multiple seasons - actual matches missed, not predicted injury risk. The headline figure - seven of every ten players in the top tier suffering a significant absence - is an out-of-sample precision measurement, not a backfit.
What we don’t publish
The 24 signals, their tiered combination, and the per-league calibration are proprietary. We publish the precision, the lift over baseline, and the per-league design principle - not the feature set itself.
The underlying data and models are proprietary. We show the validated results, not the inputs that produce them.
See it applied to your shortlist.
Every report shows the model’s output with the per-league accuracy behind it.
Order a Report