Table 5.
Final score. The final score is obtained by ensembling a plurality of models in which the best score is recorded for each sound using a technique called stacking and using the final stage model. The correct answer rate for the test data was unknown 72–93% (average 82%), and the AUC was 0.75–0.92% (average 0.83%).
| NORMAL | HARD | HIGH | INTERMITTENT | WHISTLE | MEAN | |
|---|---|---|---|---|---|---|
| accuracy | 0.753 | 0.833 | 0.729 | 0.935 | 0.855 | 0.821 |
| precision | 0.45 | 0.766 | 0.478 | 0.014 | 0.36 | 0.414 |
| recall | 0.468 | 0.862 | 0.478 | 0.5 | 0.581 | 0.578 |
| specificity | 0.836 | 0.812 | 0.817 | 0.936 | 0.885 | 0.857 |
| f1 | 0.459 | 0.811 | 0.478 | 0.028 | 0.444 | 0.444 |
| AUC | 0.759 | 0.889 | 0.795 | 0.929 | 0.845 | 0.843 |