Table 3 Performance metrics (mean ± STD) of difference ML models for PTB detection in internal validation and blinded test dataset
From: A cross-sectional study: a breathomics based pulmonary tuberculosis detection method
Data sets | Models | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | Accuracy (%) | AUC |
|---|---|---|---|---|---|---|---|
Validation (n = 295) | RF | 90.6 ± 3.1 | 90.6 ± 2.4 | 85.1 ± 3.2 | 94.3 ± 1.7 | 90.6 ± 1.7 | 0.960 ± 0.011 |
SVM | 67.7 ± 20.0 | 83.4 ± 12.6 | 74.3 ± 10.4 | 83.1 ± 7.6 | 77.6 ± 4.8 | 0.755 ± 0.061 | |
LR | 78.6 ± 4.4 | 82.0 ± 4.1 | 72.2 ± 4.7 | 86.8 ± 2.5 | 80.8 ± 3.1 | 0.856 ± 0.030 | |
XGB | 88.1 ± 3.0 | 93.6 ± 2.1 | 89.0 ± 3.2 | 93.1 ± 1.6 | 91.5 ± 1.6 | 0.969 ± 0.010 | |
DT | 76.1 ± 5.0 | 90.5 ± 2.8 | 82.6 ± 4.2 | 86.7 ± 2.4 | 85.2 ± 2.5 | 0.833 ± 0.028 | |
Test (n = 430) | RF | 90.7 ± 1.5 | 92.1 ± 1.5 | 86.9 ± 2.1 | 94.5 ± 0.8 | 91.6 ± 1.0 | 0.970 ± 0.005 |
SVM | 69.4 ± 20.4 | 83.6 ± 12.7 | 74.5 ± 9.7 | 84.3 ± 7.8 | 78.4 ± 5.2 | 0.765 ± 0.066 | |
LR | 82.5 ± 3.3 | 83.2 ± 4.0 | 74.1 ± 4.5 | 89.2 ± 1.8 | 82.9 ± 2.8 | 0.877 ± 0.021 | |
XGB | 88.1 ± 1.7 | 94.6 ± 1.2 | 90.5 ± 2.0 | 93.2 ± 0.9 | 92.2 ± 0.9 | 0.970 ± 0.004 | |
DT | 75.3 ± 4.1 | 89.4 ± 1.9 | 80.5 ± 3.0 | 86.3 ± 2.0 | 84.3 ± 1.9 | 0.824 ± 0.023 |