Generalize Then Personalize: Wearable-Based Heart Rate Forecasting Under Scarce Ruck-March Data
Organizations: University of Massachusetts Lowell, Lowell, MA, USA · U.S. Army DEVCOM Soldier Center, Natick, MA, USA
Abstract
Heart rate (HR) forecasting may provide advance information about near-future cardiovascular response, potentially supporting physiological monitoring and workload-management decisions during ruck marching. However, robust model development is constrained by the difficulty of collecting large-scale ruck-march datasets, creating a scarce-data problem in which models must generalize across participants while adapting to individuals from limited personal data. To address this problem, we benchmark population-only training (POT), personalized-from-scratch training (PFST), and population-pretrained fine-tuning (PPFT) for HR forecasting during loaded ruck marching using data from eight active-duty U.S. Army medics across thirteen sessions. LSTM, Attention-LSTM, and CNN-LSTM backbones were evaluated against a persistence baseline at 30 s and 60 s horizons. Across all neural architectures and horizons, PPFT yielded lower mean MAE, RMSE, and endpoint MAE than both POT and PFST. Participant-level paired analysis after averaging across three training seeds showed that, relative to PFST, PPFT reduced trajectory MAE by 2.22-4.30 bpm and endpoint MAE by 1.24-2.88 bpm. All six trajectory-MAE comparisons remained significant after Holm adjustment ( = 0.0469), while four of six endpoint-MAE comparisons remained significant after adjustment. At 30 s, persistence achieved the lowest MAE (3.555 1.274 bpm), while CNN-LSTM PPFT achieved the lowest RMSE (4.669 1.916 bpm); LSTM PPFT achieved the lowest endpoint MAE (5.093 2.365 bpm). At 60 s, Attention-LSTM PPFT achieved the lowest MAE (5.573 1.939 bpm) and RMSE (7.134 2.260 bpm), while CNN-LSTM PPFT achieved the lowest endpoint MAE (7.553 2.357 bpm). These results suggest that population pretraining followed by participant-specific fine-tuning may provide a practical strategy for scarce-data wearable HR forecasting.
Figures & tables
| Participant | Duration (min) | Sequences | HR Mean (bpm) | HR Range (bpm) | PL Mean | PL Range |
|---|---|---|---|---|---|---|
| Medic 1 | 15.10 | 727 | 36.0 | 109.7 | ||
| Medic 2 | 20.17 | 1031 | 51.0 | 33.3 | ||
| Medic 3 | 11.18 | 492 | 80.9 | 58.3 | ||
| Medic 4 | 13.75 | 646 | 18.0 | 46.6 | ||
| Medic 5 | 19.98 | 1020 | 46.0 | 46.5 | ||
| Medic 6 | 23.40 | 1225 | 64.4 | 180.0 |
| Model | Training Regime | MAE (bpm) | RMSE (bpm) | Endpoint MAE (bpm) |
|---|---|---|---|---|
| Baseline | Persistence | |||
| LSTM | POT | |||
| LSTM | PFST | |||
| LSTM | PPFT | |||
| Attention-LSTM | POT | |||
| Attention-LSTM | PFST |
| Model | Training Regime | MAE (bpm) | RMSE (bpm) | Endpoint MAE (bpm) |
|---|---|---|---|---|
| Baseline | Persistence | |||
| LSTM | POT | |||
| LSTM | PFST | |||
| LSTM | PPFT | |||
| Attention-LSTM | POT | |||
| Attention-LSTM | PFST |
| Participant | Persistence | LSTM | Attention-LSTM | CNN-LSTM | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| POT | PFST | PPFT | POT | PFST | PPFT | POT | PFST | PPFT | ||
| Medic 1 | 0.738 | 0.682 | 0.536 | 0.791 | 0.727 | 0.486 | 0.775 | 0.636 | 0.479 | 0.742 |
| Medic 2 | 0.127 | 0.164 | -0.493 | 0.234 | 0.176 | -0.648 | 0.207 | 0.048 | -0.467 | 0.213 |
| Medic 3 | 0.651 | 0.799 | 0.486 | 0.817 | 0.802 | 0.475 | 0.817 | 0.795 | 0.411 | 0.792 |
| Medic 4 a | -0.297 | -0.488 | -1.786 | -0.249 | -1.453 | -0.974 | -0.330 | -0.615 | -1.116 | -0.237 |
| Medic 5 | 0.583 | 0.416 | -0.940 | 0.360 | 0.386 | -1.466 | 0.343 | 0.323 | -1.237 | 0.310 |
| Model | Fine-Tuning Approach | MAE (bpm) | RMSE (bpm) | Endpoint MAE (bpm) |
|---|---|---|---|---|
| LSTM | Direct PPFT | |||
| LSTM | Decomposition PPFT | |||
| Attention-LSTM | Direct PPFT | |||
| Attention-LSTM | Decomposition PPFT | |||
| CNN-LSTM | Direct PPFT | |||
| CNN-LSTM | Decomposition PPFT |
Heart Rate Estimation
Heart rate estimation recovers beats per minute from wearable, contact, or remote physiological signals.
See all 34 papers