Age-segmented projection
Three-Axis Prediction v2
Dedicated U24 and Senior models predicting value, performance, and minutes two years forward across 31 leagues.
Age-segmented projections that explain 87% of variance in U24 on-pitch development (R² 0.872) and rank future valuations with Spearman 0.91 correlation - trained on 166,865 pairs with zero temporal leakage.
What it does
Three-Axis v2 splits players at age 24 into dedicated sub-models. Young players have fundamentally different development dynamics - rapid improvement, volatile minutes, steep value curves. Training separate models captures these patterns without contamination from the stable senior population. The U24 model explains 87% of variance in performance growth (R² 0.872), while the Senior model excels at valuation ranking (Spearman 0.923). Combined output: 1,243 breakout candidates, 1,185 high-value risers, and 501 elite-trajectory players scored daily.
Used in: Cross-model shortlisting, breakout detection, and transfer target ranking.
Why it’s defensible
What makes this proprietary.
Age segmentation as structural advantage
A global model averages across age cohorts with fundamentally different dynamics. By training dedicated U24 and Senior models (split at 24), we capture breakout trajectories that a pooled model smooths away. U24 Performance Δ jumps from R² 0.51 (global) to 0.87 (segmented) - a 70% relative improvement from architecture alone.
No-leakage evaluation design
Every metric is from 5-fold cross-validation with strict temporal controls: 2-year horizon for valuations, minimum 180-day gap for performance and minutes Δ pairs. No future information leaks into training. The numbers reflect genuine forward predictive power.
Scale and league coverage
Trained on 166,865 pairs across 31 leagues in a 4-tier hierarchy (top-5 European to emerging markets). League tier is an explicit feature, so the model understands that a Liga NOS breakout and a Bundesliga breakout have different base rates and trajectory shapes.
How we validate it
Each axis validated independently via 5-fold CV with no temporal leakage. Valuation: Spearman 0.91 (global), 0.923 (senior), 0.866 (U24). Performance Δ: R² 0.872 (U24), 0.125 (senior) - senior players are near-ceiling, so low R² is expected and honest. Minutes Δ: Spearman 0.65 across all segments. The model uses HistGradientBoosting as primary learner with a Ridge-based stacking ensemble.
What we don’t publish
The age-segmentation boundaries, per-segment feature engineering, 4-tier league hierarchy weights, and stacking architecture are proprietary. We publish validated accuracy per axis and segment so the result can be trusted without exposing how it is produced.
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