Table 2 Performance of the hierarchical classifier over the test set (shallow neural network + shallow neural network in the case of 3D RI tomograms and logistic regression + linear discriminant in the case of 2D QPMs).
Metric | 3D RI Tomogram | 2D QPM | ||||
|---|---|---|---|---|---|---|
MC | NB | OC | MC | NB | OC | |
Accuracy | 95.2 | 76.0 | ||||
True positive rate (sensitivity or recall) | 95.7 | 93.4 | 98.3 | 63.8 | 77.9 | 81.7 |
True negative rate (specificity) | 97.8 | 97.2 | 97.6 | 93.4 | 86.9 | 82.8 |
Positive predictive value (precision) | 91.8 | 97.4 | 93.7 | 71.4 | 89.2 | 62.8 |
Negative predictive value | 98.9 | 92.9 | 99.4 | 90.9 | 77.5 | 92.7 |
Balanced accuracy | 96.8 | 95.3 | 98.0 | 78.6 | 82.4 | 82.2 |
F1 score | 93.8 | 95.4 | 95.9 | 67.4 | 82.3 | 71.0 |
Matthews correlation Coefficient | 92.1 | 90.5 | 94.5 | 59.7 | 64.7 | 59.9 |
Fowlkes–Mallows index | 93.8 | 95.4 | 96.0 | 67.5 | 82.4 | 71.6 |
