Table 3.
Results individually per patient and averages in predicting glucose levels with a 30- and 60-min prediction horizon respectively
| 30-min horizon | 60-min horizon | |||
|---|---|---|---|---|
| Patient ID | LSTM | t0 | LSTM | t0 |
| 559 | 0.178 ± 0.003 | 0.224 | 0.331 ± 0.003 | 0.386 |
| 570 | 0.105 ± 0.002 | 0.141 | 0.195 ± 0.004 | 0.244 |
| 588 | 0.177 ± 0.002 | 0.214 | 0.291 ± 0.002 | 0.349 |
| 563 | 0.176 ± 0.002 | 0.222 | 0.293 ± 0.002 | 0.360 |
| 575 | 0.224 ± 0.004 | 0.272 | 0.389 ± 0.005 | 0.434 |
| 591 | 0.256 ± 0.003 | 0.299 | 0.396 ± 0.003 | 0.478 |
| μ | 0.186 | 0.229 | 0.316 | 0.375 |
| σ | ± 0.047 | ± 0.050 | ± 0.068 | ± 0.073 |
The table shows the surveillance error (SE) of the LSTM model trained with NLL. t0 refers to the naive baseline of predicting the last value