Table 1.

Evaluation of AMR classifiers for first-line and second-line anti-TB drugs (our ML methods VS the rule-based one Mykrobe predictor).

Methods	Precision (%)	Sensitivity (%)	Specificity (%)	Accuracy (%)	F1 (%)	G-mean (%)
First-line
INH
RF	95.2	98.7	91.7	96.1	97.0	95.1
LR	94.3	99.2	90.0	95.7	96.7	94.5
CNN	95.5	99.0	90.5	96.2	97.2	94.7
Mykrobe	92.9	99.2	95.3	96.2	95.9	97.2
PZA
RF	92.3	96.4	53.6	90.1	94.3	71.9
LR	93.0	95.5	58.0	90.0	94.2	74.4
CNN	93.2	96.5	56.1	90.5	94.8	73.6
Mykrobe	91.1	95.1	60.9	87.3	93.1	76.1
RIF
RF	94.3	97.0	88.6	94.1	95.6	92.7
LR	93.5	98.3	86.6	94.3	95.8	92.3
CNN	94.4	98.1	87.5	94.6	96.2	92.7
Mykrobe	92.4	95.0	92.3	92.5	93.7	93.6
EMB
RF	92.9	94.4	70.8	89.7	93.6	81.8
LR	93.1	93.4	72.1	89.2	93.3	82.1
CNN	93.1	94.5	71.7	90.0	93.8	82.3
Mykrobe	92.2	72.4	85.3	76.3	81.1	78.6
Second-line
AMK
RF	99.1	99.7	80.9	98.9	99.4	89.8
LR	99.2	99.9	82.8	99.2	99.5	91.0
CNN	99.2	100	82.9	99.2	99.6	91.1
Mykrobe	99.1	100	81.9	99.2	99.5	90.5
CM
RF	96.0	98.7	48.0	94.9	97.3	68.8
LR	96.1	99.9	49.0	96.1	98.0	70.0
CNN	96.1	99.9	49.0	96.1	98.0	70.0
Mykrobe	96.0	99.9	47.8	96.1	97.8	69.1
KM
RF	94.0	93.9	59.0	89.4	93.9	74.4
LR	94.5	95.1	63.0	91.0	94.8	77.4
CNN	93.0	98.4	49.8	92.1	95.6	70.0
Mykrobe	90.9	99.5	31.9	91.0	95.0	56.4
OFX
RF	97.5	98.5	64.5	96.4	98.0	79.7
LR	97.6	99.1	65.2	96.9	98.3	80.4
CNN	97.5	98.9	63.4	96.6	98.2	79.2
Mykrobe	98.5	95.9	78.1	94.8	97.1	86.6