This file contains the GAM 'CategorisationGAM.rdata' from Palmer et al 2017 http://asa.scitation.org/doi/10.1121/1.4996000
Input:
MedICI- median inter-click-interval of the C-POD click train
NCycles- median number of cycles for the clicks in each click train
MeanFreq- Mean frequency of the clicks in each C-POD click train
Output:
Speciesid- the logit probability that each click train was 'frequency banded' i.e. contained spectral peaks and notches consistant with whitebeaked/risso's off-axis click trains. Since our study assumed that ESC clcik trains must be either broadband 'broadband' or 'frequency banded' the model is binary. To estimate 'broadband' probability subtract inv.logit() of the model output from 1.
Example R code provided for running GAM categorisation and likelihood ratio test. Assumes C-POD "click details" exported and pre-processed for predictors and saved as CSV.
Post Publication Work.docx includes post analysis that we've been able to do since commensing this work in 2013/2014 and includes out-of-sample performance as well as the relationship between the number of click trains in the encounter and whether or not the categorisation system correctly classified the train.