Table 3.
Results of hypoglycemia prediction (Mean ± SD) evaluated on 12 clinical T1D subject.
| Prediction horizons | 15 min | 30 min | 45 min | 60 min | 60 min (Baseline44) |
|---|---|---|---|---|---|
| Accuracy (%) | 98.03 ± 1.03 | 94.96 ± 2.92 | 91.97 ± 4.22 | 88.58 ± 6.53 | 91.89 ± 5.23 |
| Sensitivity (%) | 84.15 ± 4.20 | 76.08 ± 5.88 | 72.07 ± 4.45 | 70.30 ± 12.84 | 56.95 ± 19.24** |
| Specificity (%) | 98.72 ± 0.75 | 96.42 ± 2.48 | 93.99 ± 4.08 | 90.09 ± 8.21 | 94.87 ± 3.67 |
| Precision (%) | 78.91 ± 4.31 | 65.65 ± 5.31 | 58.23 ± 10.21 | 56.20 ± 10.43 | 55.28 ± 17.38 |
| MCC score | 0.80 ± 0.04 | 0.68 ± 0.05 | 0.60 ± 0.06 | 0.56 ± 0.07 | 0.51 ± 0.12 |
| MD (mg/dL) | 10.28 ± 3.80 | 19.18 ± 7.00 | 26.30 ± 9.28 | 28.63 ± 11.00 | 41.92 ± 14.60** |
MCC Matthews correlation coefficient, MD mean deviation from true glucose levels for missed predicted hypoglycemic events. The best performance of the baseline methods (ARMA44) is presented.
The significance is indicated as **p < 0.01.