Predictive Models

Age-only and age + best wearable predictor ยท LOO cross-validated ยท 2026-11-21
โš ๏ธ Statistical disclaimer โ€” These are simple OLS models with n=5โ€“15 data points. Results are directional and exploratory, not clinical. In-sample Rยฒ is optimistic; LOO Rยฒ (leave-one-out cross-validated) is the honest estimate of predictive power. Where LOO Rยฒ is negative, the model predicts worse than the mean โ€” treat it as noise. Max 2 predictors (age + 1 wearable) to avoid overfitting. Prediction intervals are ยฑ1 LOO-MAE, not formal confidence intervals.
biomarkers modelled
15
with wearable predictor
15
Andy's age today
37
HRV 30d avg today
โ€”
Sleep hrs 30d today
โ€”
Strain 30d today
โ€”
LOO Rยฒ by biomarker
Model B where a wearable predictor qualified, otherwise Model A. Bars are cut off at โˆ’1; below zero the model predicts worse than the mean.