Garson´s weight method or Connection Weight approch proposed by Olden et al., (2004)?
Garson, G. D. (1991). Interpreting neural-network connection weights. AI expert, 6(4), 46-51.
Olden, J. D., Joy, M. K., & Death, R. G. (2004). An accurate comparison of methods for quantifying variable importance in artificial neural networks using simulated data. Ecological Modelling, 178(3-4), 389-397.
I suggest training the model the number of times that corresponds to the number of inputs you have. On each occasion you remove an input and evaluate the performance of the model compared to the model with all the inputs. We applied this in this work (Article Neural based contingent valuation of road traffic noise
) and it was quite adequate (look for the content related to table 6).
I suggest training the model the number of times that corresponds to the number of inputs you have. On each occasion you remove an input and evaluate the performance of the model compared to the model with all the inputs. We applied this in this work (Article Neural based contingent valuation of road traffic noise
) and it was quite adequate (look for the content related to table 6).
Though a bit complicated to implement in case of neural networks with large number of hidden neurons and inputs, for simple ANN models, Gandomi et al. (2013)'s approach may be used. I attach below the reference (check section 5 'Sensitivity analysis' of the article). Thanks.
Reference:
Gandomi, A.H.; Yun, G.J.; Alavi, A.H.: An evolutionary approach for modeling of shear strength of RC deep beams. Mater. Struct. (2013) doi:10.1617/s11527-013-0039-z
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity