244
M. Gevrey . S. Lek . T. Oberdorff
The second is a c1assification of the independent variable in increasing order of
influence. Moreover, when a variable is more dominant than another, the partial
derivatives obtained for the other are near to zero.
The profile method, because it presents all the variables at the same scale
permits a general view of how the independent variables influence the output.
When a variable is dominant, a curve with a large range is represented for this
variable while the other curves are "crushed".
These methods are able to provide explanations for the model. The most
influential variable is known, as is the order of influence of all the variables, and
the way that these variables intervene in the model.
12.6
Conclusions
The results obtained with both methods match c10sely with the previous results.
The predictive power of ANNs has often been demonstrated, and this new study
puts to the fore their explicative power which is very interesting in ecological
research.
This artic1e paves the way forward for broad research concerning the
contribution of the input variables in ANN's, firstly by the use of other databases
to test the methods, secondly by the discovery of new methods and finally by the
investigation of other existing methods.
References
Balls GR, Palmer-Brown D, Sanders GE (1996) Investigating microclimatic influences on
ozone injury in clover (Trifolium subterraneum) using artificial neural networks. New
Phytol. 132,271-280
Brey T, Jarre-Teichmann A, Borlieh 0 (1996) Artificial neural network versus multiple
linear regression: predicting P/B ratios from empirical data. Marine Ecology Progress
Series, 140,251-256
Clair TA, Ehrman JM (1998) Using neural networks to assess the influence of changing
seasonal climates in modifying discharge, dissolved organie carbon, and nitrogen
export in eastem Canadian rivers. Water Resources Research, 34(3), 447-455
Comuet JM, Aulagnier S, Lek S, Franck P, Solignac M (1996) Classifying individuals
among intra-specific taxa using microsatellite data and neural networks. C. R. Acad.
Sei. Paris, Sciences de la vie 319,1167-77
Dimopoulos Y, Bourret P, Lek S (1995) Use of some sensitivity criteria for choosing
networks with good generalization ability. Neural Process. Lett. 2(6), 1-4
Dimopoulos Y, Chronopoulos J, Chronopoulou-Sereli A, Lek S (1999) Neural network
models to study relationships between lead concentration in grasses and permanent
urban descriptors in Athens city (Greece). Ecological Modelling 120, 157-165
Efron B (1983) Estimating the error rate of aprediction rule: improvement on crossvalidation. J. Am. Stat. Assoe. 78, 316-330
M. Gevrey . S. Lek . T. Oberdorff
The second is a c1assification of the independent variable in increasing order of
influence. Moreover, when a variable is more dominant than another, the partial
derivatives obtained for the other are near to zero.
The profile method, because it presents all the variables at the same scale
permits a general view of how the independent variables influence the output.
When a variable is dominant, a curve with a large range is represented for this
variable while the other curves are "crushed".
These methods are able to provide explanations for the model. The most
influential variable is known, as is the order of influence of all the variables, and
the way that these variables intervene in the model.
12.6
Conclusions
The results obtained with both methods match c10sely with the previous results.
The predictive power of ANNs has often been demonstrated, and this new study
puts to the fore their explicative power which is very interesting in ecological
research.
This artic1e paves the way forward for broad research concerning the
contribution of the input variables in ANN's, firstly by the use of other databases
to test the methods, secondly by the discovery of new methods and finally by the
investigation of other existing methods.
References
Balls GR, Palmer-Brown D, Sanders GE (1996) Investigating microclimatic influences on
ozone injury in clover (Trifolium subterraneum) using artificial neural networks. New
Phytol. 132,271-280
Brey T, Jarre-Teichmann A, Borlieh 0 (1996) Artificial neural network versus multiple
linear regression: predicting P/B ratios from empirical data. Marine Ecology Progress
Series, 140,251-256
Clair TA, Ehrman JM (1998) Using neural networks to assess the influence of changing
seasonal climates in modifying discharge, dissolved organie carbon, and nitrogen
export in eastem Canadian rivers. Water Resources Research, 34(3), 447-455
Comuet JM, Aulagnier S, Lek S, Franck P, Solignac M (1996) Classifying individuals
among intra-specific taxa using microsatellite data and neural networks. C. R. Acad.
Sei. Paris, Sciences de la vie 319,1167-77
Dimopoulos Y, Bourret P, Lek S (1995) Use of some sensitivity criteria for choosing
networks with good generalization ability. Neural Process. Lett. 2(6), 1-4
Dimopoulos Y, Chronopoulos J, Chronopoulou-Sereli A, Lek S (1999) Neural network
models to study relationships between lead concentration in grasses and permanent
urban descriptors in Athens city (Greece). Ecological Modelling 120, 157-165
Efron B (1983) Estimating the error rate of aprediction rule: improvement on crossvalidation. J. Am. Stat. Assoe. 78, 316-330