Table 1.
Performance of the best machine learning classifier and DA considering INH, EMB, RIF, PZA, MDR-TB, and FDR-TB.
| DA | Best method | ||||||
|---|---|---|---|---|---|---|---|
| Drugs | Sensitivity | Specificity | AUC | Feature set + Classifier | Sensitivity | Specificity | AUC |
| INH | 91.15 ± 1.19 | 98.96 ± 0.25 | 95.05 ± 0.60 | F3 + MLRF | 93.76* ± 0.80 | 97.79 ± 0.35 | 96.01* ± 0.47 |
| EMB | 85.10 ± 1.79 | 94.91 ± 0.38 | 90.00 ± 0.97 | F1 + MLRF | 91.75* ± 1.81 | 91.58* ± 0.77 | 91.70* ± 0.75 |
| RIF | 91.52 ± 1.34 | 98.68 ± 0.21 | 95.10 ± 0.65 | F3 + MLRF | 93.16* ± 0.80 | 98.02 ± 0.32 | 96.00* ± 0.40 |
| PZA | 43.21 ± 2.72 | 98.58 ± 0.23 | 70.89 ± 1.35 | F1 + SLRF | 87.27* ± 1.74 | 90.71* ± 0.72 | 88.99* ± 0.84 |
| FDR-TB | 37.34 ± 3.97 | 98.59 ± 0.22 | 67.96 ± 1.99 | F1 + MLRF | 87.58* ± 2.79 | 92.98* ± 0.45 | 90.28* ± 1.23 |
| MDR-TB | 89.84 ± 1.34 | 99.12 ± 0.178 | 94.48 ± 0.69 | F3 + MLRF | 93.70* ± 0.76 | 97.45 ± 0.36 | 95.58* ± 0.41 |
Sensitivity, specificity and AUC (mean ± standard error) were reported. The Wilcoxon signed-rank test was used to calculate the p-value of each method compared with the DA and
*
p < 0.01 vs. DA.