Table 1.
Model comparison for predicting fine-grained VO2max with the Fenland I cohort.
| Features | Models | Evaluation Metrics [95% CI] | N (train/test set) | ||
|---|---|---|---|---|---|
| R2 | Corr | RMSE | |||
| Anthropometrics | Linear | ||||
| Age/Sex/Weight/BMI/Height | 0.359 [0.329–0.388] | 0.600 [0.577–0.623] | 4.051 [3.947–4.170] | ||
| Resting Heart Rate | |||||
| RHR (sensor-derived) | 0.373 [0.342–0.403] | 0.612 [0.587–0.638] | 4.007 [3.885–4.113] | 11059 (8384/2675) | |
| Anthropometrics + RHR | |||||
| Age/Sex/Weight/BMI/Height/RHR | 0.610 [0.582–0.634] | 0.781 [0.764–0.796] | 3.159 [3.051–3.272] | ||
| Wearable Sensors + RHR + Anthro. | |||||
| Acceleration/HR/HRV/MVPA Age/Sex/Weight/BMI/Height/RHR | 0.658 [0.623–0.685] | 0.812 [0.792–0.828] | 2.956 [2.830–3.082] | ||
| Dense | 0.671 [0.649–0.691] | 0.821 [0.806–0.835] | 2.902 [2.806–3.002] | ||
Comparison between linear regression and a dense neural network trained on combinations of antrhopometrics, common biomarkers (RHR), and passively collected data over a week (wearablesensors). Best performance in bold. The units of VO2max are measured in mlO2/min/kg. Results reported from the testing set.