Table 2.
CNN and LSTM accuracy in estimating energy expenditure considering the LOOCV algorithm and the test dataset.
| Model | Hyperparameters | MSE a | RMSE b | NMSE c | SCC d | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| LOOCV | Test | LOOCV | Test | LOOCV | Test | LOOCV | Test | |||
| LSTM | Neurons | 150 | 0.25 (0.22) |
N.A. | 0.45 (0.22) |
N.A. | 0.67 (0.23) |
N.A. | 0.86 (0.035) |
N.A. |
| Layers | 1 | |||||||||
| Batch size | 76 | |||||||||
| FC Layer | 1 | |||||||||
| CNN | Filter size | 10 | 0.14 (0.10) |
0.19 | 0.36 (0.13) |
0.44 | 0.79 (0.12) |
0.71 | 0.87 (0.043) |
0.87 |
| No. filters | 2 × 8 | |||||||||
| Batch size | 22,800 | |||||||||
| FC Layer | 10 | |||||||||
a Mean-Square Error; b Root-Mean-Square Error; c Normalized Mean-Square Error; d Spearman’s Correlation Coefficient.