Table 2. Sensitivity (Recall), Specificity, Precision, and F1 Score of the eCOV Modela.
| Message category | No. (%) of patients | Sensitivity, % | Specificity, % | Precision, % | F1 score, % |
|---|---|---|---|---|---|
| COVID-19 other | 148 (5.1) | 85 | 100 | 90 | 87 |
| COVID-19 positive | 307 (10.6) | 96 | 99 | 93 | 95 |
| Non–COVID-19 | 2452 (84.4) | 100 | 99 | 100 | 100 |
| Unweighted mean | NA | 94 | 99 | 94 | 94 |
| Weighted mean | NA | 99 | 99 | 99 | 99 |
Abbreviation: NA, not applicable.
a
Class-specific sensitivity was calculated as true positive divided by true positive plus false negative [(TP)/(TP + FN)] in a one-vs-rest approach; class-specific specificity, true negative divided by true negative plus false positive [TN/(TN + FP)] in a one-vs-rest approach; class-specific precision, TP/(TP + FP) in a one-vs-rest approach; and class-specific F1 score, 2TP/(2TP + FP + FN) in a one-vs-rest approach. The unweighted mean (macro mean) of each value was the arithmetic mean. The weighted mean of each value included the number of patients for each class into calculation.