Small to train, small to test: Dealing with low sample size in model evaluation

Details

Ressource 1Request a copy Under embargo until 24/04/2025.
UNIL restricted access
State: Public
Version: Final published version
License: All rights reserved
Serval ID
serval:BIB_DDA1566B06B4
Type
Article: article from journal or magazin.
Collection
Publications
Institution
Title
Small to train, small to test: Dealing with low sample size in model evaluation
Journal
Ecological Informatics
Author(s)
Collart Flavien, Guisan Antoine
ISSN
1574-9541
Publication state
Published
Issued date
07/2023
Peer-reviewed
Oui
Volume
75
Pages
102106
Language
english
Abstract
Sample size is a key issue in species distribution modelling. While many studies focused on the relevance of sample size for model calibration, the importance of the size of the dataset used for model evaluation has received much less attention. Here, we highlight two previously published approaches to address the problem, and which are relatively simple to implement: the pooling evaluation and the implementation of null models. We discuss the importance of these or other potential approaches that are critical for model evaluation in rare species, which represent the bulk of biodiversity, and for which accurate models are most necessary in a conservation context.
Keywords
Species distribution model, Ecological niche, Evaluation, Sample size, Rare species, Conservation
Web of science
Funding(s)
Swiss National Science Foundation / 197777
Create date
19/04/2023 12:51
Last modification date
26/08/2023 5:52
Usage data