Skip to main content

Table 2 The overall model accuracy and mean F1 statistic (harmonic mean of the precision and recall) for each 10 Hz model using different strategies to build the random forest model. *SD not available

From: Limitations of using surrogates for behaviour classification of accelerometer data: refining methods using random forest models in Caprids

Model Pygmy goat Alpine ibex
Accuracy ± SD (%) F1 statistic ± SD Accuracy ± SD (%) F1 statistic ± SD
Random split train and test data 97.8 0.96 ± 0.02 98.6 0.96 ± 0.01
with balanced observations 97.6 0.98 ± 0.02 98.6 0.99 ± 0.01
with reduced behaviours 98.2 0.97 ± 0.02 98.7 0.97 ± 0.01
Data split per individual train and test data 57.8 ± 5.4 0.34 ± 0.03 65.5 ± 5.2 0.40 ± 0.02
with balanced observations 42.1 ± 6.0 0.38 ± 0.05 47.7 ± 7.2 0.43 ± 0.05
with reduced behaviours 59.2 ± 5.7 0.43 ± 0.05 68.6 ± 5.1 0.51 ± 0.02