Table 2.
Model | Sen (%,SD) | Spe (%,SD) | F1 score (SD) | AUC (SD) | NH (SD) | TPe (SD) | FAR (SD) | DT (min,SD) |
RF | 89.6 | 91.3 | 0.543 | 0.966 | 36.4 | 30.2 | 0.704 | 25.5 |
(2.78) | (2.03) | (0.053) | (0.007) | (11.0) | (8.42) | (0.035) | (1.97) | |
SVM | 93.3 | 88.2 | 0.487 | 0.967 | 36.4 | 29.2 | 0.777 | 25.8 |
-LN | (1.70) | (2.83) | (0.046) | (0.007) | (11.0) | (8.30) | (0.034) | (2.12) |
SVM | 89.9 | 88.8 | 0.487 | 0.952 | 36.4 | 29.4 | 0.760 | 25.2 |
-RBF | (8.65) | (2.96) | (0.062) | (0.014) | (11.0) | (9.20) | (0.038) | (3.22) |
KNN | 88.5 | 89.4 | 0.492 | 0.917 | 36.4 | 29.6 | 0.779 | 25.8 |
(1.93) | (2.09) | (0.054) | (0.012) | (11.0) | (8.73) | (0.038) | (3.76) | |
LR | 93.6 | 87.9 | 0.484 | 0.967 | 36.4 | 29.6 | 0.772 | 25.0 |
(2.25) | (2.95) | (0.047) | (0.007) | (11.0) | (8.71) | (0.037) | (2.87) |
With the 5-fold cross-subject validation, average metrics were calculated using Eq 6, 7, 9, and 10 on test setq,q=1,2,3,4,5. Since there should be at least two consecutive predictions of a hypoglycemia alert value to make an alarm, we excluded hypoglycemic events occurring immediately after meals. Abbreviation: RF, random forest; SVM-LN, support vector machine with a linear kernel; SVM-RBF, support vector machine with a radial basis function; KNN, K-nearest neighbor; LR, logistic regression; SD, standard deviation; Sen, sensitivity; Spe, specificity; AUC, the area under the ROC curve; NH, the number of hypoglycemic events; FAR, false alarm rate; DT, detection time.