Table 2. Cross-validation (training) and out-of-sample (testing) accuracy for gradient boosting models (GBM) trained across a range of epochs using two datasets for all behaviours (ALL) and for behaviour in water (Water).
| Behaviour | Hz | Epochs | Cross-validation accuracy | Out-of-sample accuracy | Kappa |
|---|---|---|---|---|---|
| ALL | 1 | 7 | 78.3% | 72.1% | 71.1% |
| ALL | 1 | 15 | 79.7% | 86.8% | 73.6% |
| ALL | 1 | 21 | 80.8% | 89.5% | 73.6% |
| Water | 20 | 13 | 63.2% | 67.6% | 68.6% |
| Water | 20 | 25 | 72.4% | 69.1% | 63.2% |
| Water | 20 | 75 | 82.7% | 75.6% | 76.9% |