pone.0259156.g003.tif (999.95 kB)
Comparison of class prevalence.
figure
posted on 2021-10-29, 17:32 authored by Edward J. Gregr, Dana R. Haggarty, Sarah C. Davies, Cole Fields, Joanne LessardObserved class prevalence in the build testing partition (orange) compared to predictions from the weighted (yellow) and unweighted (blue) random forest models across regions. Weighting tends to yield class prevalence closer to that observed in the build data (note that training and testing partitions have the same prevalence).
History
Usage metrics
Categories
Keywords
true skill statisticrandom forest classificationoverlook shallow watersaccurately representing heterogeneitytest model predictionscanada &# 8217known spatial featuresindependent data evaluationbuild data wellmodel performance influencedassessing model stationaritypredict marine substratediv >< palso urgently neededmodel performancemodel fitstationarity acrossmarine resourcescoastal substratecanada ’spatial regionsspatial categorieswork showstimely alternativetime consumingstudy areastandard testsscores rangingpreviously achievedpredictive powerpacific shelfmodels fitindicating nonimprove understandinggrain sizefoundational rolefocus largelyerror metricscurrent approachescountries workbottom typealso testedalmost 200000 km
Licence
Exports
RefWorks
BibTeX
Ref. manager
Endnote
DataCite
NLM
DC