Fig. 2
From: A comparison of statistical methods for deriving occupancy estimates from machine learning outputs
Comparison of modelled occupancy estimates produced by: (a) standard occupancy model, (b) binary false-positive model and extensions, (c) frequency false positive occupancy model using counts of detections, and (d) false positive occupancy models using raw classifier scores. Where applicable models were run with the addition of two different types of verified data from each site: 10 randomly selected files, and 10 files with the highest classifier score. A threshold of 0.77 was used for binary and frequency false-positive models. Results were removed for variations that did not converge (e.g. Kéry model with randomly verified data, and without verified data). Bars denote 95% confidence intervals and credible intervals for maximum likelihood and Bayesian approaches respectively. The dashed vertical line represents the sample-at-hand occupancy estimate.
