Comparative of learning and predictive models in ecology
Sovan Lek, Muriel Gevrey, Gael Grenouillet
Last modified: 2008-09-13
Abstract
This paper aims to compare the powers of advanced modelling learning methods to predict and explain ecological data.
Two datasets were used to predict qualitative and quantitative outputs (biological variables). The input variables are a set of dozen environmental quantitative and qualitative descriptors.
Modern modelling techniques were tested: artificial neural networks algorithms, generalized linear and additive models, classifications and regression trees, random forest and boosted trees, support vector machines... For each method, predictive results were analysed and discussed. The importance of each variable for each model was also discussed.
Two datasets were used to predict qualitative and quantitative outputs (biological variables). The input variables are a set of dozen environmental quantitative and qualitative descriptors.
Modern modelling techniques were tested: artificial neural networks algorithms, generalized linear and additive models, classifications and regression trees, random forest and boosted trees, support vector machines... For each method, predictive results were analysed and discussed. The importance of each variable for each model was also discussed.