Validation of the performance of the random forest model for activity classification

We applied the trained random forest models to the 50% data withheld for testing to evaluate its performance in classifying bat activity. Similarly, we applied random forest models to the bird and human activity dataset after calculating the same predictor variables as for the bats. We first calculated the true positive rate (TPR) as the ratio of correctly identified incidents by comparing the observed data with the activity class attributed by the trained random forest models for all dataset types (i.e., human activity dataset, woodpecker, and bat video sequences). Next, we calculated the models’ F-score (F1) using the Caret package (Kuhn, 2008). Data and code are stored here.