Premium
Use of decision tree and artificial neural network approaches to model presence/absence of Telestes muticellus in piedmont (North‐Western Italy)
Author(s) -
Tirelli Tina,
Pessani Daniela
Publication year - 2009
Publication title -
river research and applications
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.679
H-Index - 94
eISSN - 1535-1467
pISSN - 1535-1459
DOI - 10.1002/rra.1199
Subject(s) - decision tree , artificial neural network , decision tree model , tree (set theory) , computer science , endangered species , artificial intelligence , machine learning , ecology , mathematics , biology , habitat , mathematical analysis
In Piedmont (Italy) the impact of human beings is causing some deep environmental changes in freshwaters and their inhabitants, so much so that we need to develop some practical tools for immediate use in providing accurate ecological assessments of the freshwater system and of the conditions of the species living there, one of which is Telestes muticellus , an endangered Cyprinidae found in the western Alps and the central Apennines in Italy. We aimed to help manage this species by assessing its presence using two types of data‐mining approaches—decision‐tree models and artificial neural networks. We built models using 10 environmental input variables to classify sites as positive or negative for the species. The unpruned decision tree models classified a high percentage of instances correctly and made accurate predictions, as did the post‐pruned tree models. The post‐pruned methods yielded simpler trees and therefore clearer models. Generally, the artificial neural networks (ANN) performed better than the decision tree models, except in the case of Cohen's k. We used the sensitivity analysis technique to understand which inputs are the most important ones for building the ANN model we obtained. Copyright © 2008 John Wiley & Sons, Ltd.