Table 3 Confusion matrices for the (a) standard Random Forest model, (b) BANTER call classifiers (Detectors 2, 3, 4, and 5), and (c) BANTER event classifier. The total number of correct and incorrect classifications are shown between the groups identified a-priori (rows) and predictions by the classifier (columns). Out-of-bag (OOB) percent correct classification rates (95% confidence interval) demonstrate model performance. Binomial test p-values using a prior probability of 0.5 are provided where an asterisk (*) represents statistical significance (p < 0.05).
From: Acoustic differentiation and classification of wild belugas and narwhals using echolocation clicks
Beluga | Narwhal | Correct classification (95% CI) | p-value | |
|---|---|---|---|---|
(a) Standard Random Forest model | ||||
Beluga | 18 | 1 | 94.7% (74.0–99.9%) | < 0.001* |
Narwhal | 5 | 57 | 91.9% (82.2–97.3%) | < 0.001* |
Overall | 92.6% (84.6–97.2%) | < 0.001* | ||
(b) BANTER call classifier | ||||
Detector 2 | ||||
Beluga | 351 | 31 | 91.9% (88.7–94.4%) | < 0.001* |
Narwhal | 886 | 5022 | 85.0% (84.1–85.9%) | < 0.001* |
Overall | 85.4% (84.5–86.3%) | < 0.001* | ||
Detector 3 | ||||
Beluga | 1007 | 122 | 89.2% (87.2–90.9%) | < 0.001* |
Narwhal | 540 | 1632 | 75.1% (73.3–76.9%) | < 0.001* |
Overall | 80.0% (78.5–81.3%) | < 0.001* | ||
Detector 4 | ||||
Beluga | 710 | 136 | 83.9% (81.3–86.3%) | < 0.001* |
Narwhal | 121 | 449 | 78.8% (75.2–82.1%) | < 0.001* |
Overall | 81.9% (79.7–83.8%) | < 0.001* | ||
Detector 5 | ||||
Beluga | 131 | 39 | 77.1% (70.0–83.1%) | < 0.001* |
Narwhal | 22 | 72 | 76.6% (66.7–84.7%) | < 0.001* |
Overall | 76.9% (71.3–81.8%) | < 0.001* | ||
(c) BANTER event classifier | ||||
Beluga | 19 | 0 | 100% (82.4–100%) | < 0.001* |
Narwhal | 2 | 60 | 96.8% (88.8–99.6%) | < 0.001* |
Overall | 97.5% (91.4–99.7%) | < 0.001* | ||
