Table 1 Hypotheses tested in this work
From: A benchmark for computational analysis of animal behavior, using animal-borne tags
Hypothesis | Hypothesis confirmed? |
|---|---|
(H1) Deep neural network-based approaches will outperform classical approaches based on hand-chosen summary statistics. | Yes, for the approaches we tested |
(H2) Self-supervised pre-training using human accelerometer data will improve classification performance. | Partly |
(H3) Self-supervised pre-training using human accelerometer data will improve classification performance when the amount of training data is reduced by a factor of four by removing individuals. | Yes |
(H4) In terms of a single model’s predictive performance, there is minimal improvement in some behavior classes when increasing the amount of training data by four times by adding individuals. | Yes |