Abstract
Artificial neural networks are good non-linear function approximators but their multi-layer, non-linear form gives little immediate indication of the features they have learnt. Several methods are put forward in this paper that reduce the complexity of the network or give simplified equations that are easier to interpret. Relative weight analysis and equation synthesis are summarised while correlated activity pruning is introduced and explained in detail. The former techniques use the weights of a trained network to assign importance to inputs or groups of inputs. The latter algorithm reduces complexity of a network by merging hidden units that have correlated activations. This procedure also allows the relationship between detected features to be evaluated. Data from pollutant impact studies are used but the techniques developed are applicable to many scientific data modelling environments.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
Benton J, Fuhrer J, Gimeno BS, Skarby L, Balls G, Palmer-Brown D, Roadknight C and Sanders G.1996. ICP-Crops and critical levels of ozone for injury development. Exceedences of Critical Loads and Levels. In Exceedences of Critical Loads and Levels. Eds. M. Knoflacher, J. Schneider and G. Soja. Umweltbundesamt (Federal Environment Agency) Wein, Austria. 97–112.
Benton J, Fuhrer J, Gimeno BS, Skarby L, Balls G, Roadknight C and Sanders-Mills G. 1996. The critical level of ozone for visible injury on crops and natural vegetation (ICP-Crops). In: Critical Levels of Ozone in Europe: Testing and Finalising the Concepts. UN ECE Workshop Report. Eds. L Karenlampi and L Skarby. University of Kuopio, Finland. 44–57.
Heck, WW, Taylor O. and Tingley DT. (eds) 1988. Assessment of Crop Loss from Air Pollutants. Elsevier Applied Science, New York.
Funahashi K. 1989. On the approximate realization of continuous mappings by neural networks. Neural Networks, 2, 183–192.
Burke HB, Hoang A and Rosen DB. 1995. Survival function estimates in cancer using artificial neural networks. Proceedings of WCNN. Vol II. p.748–749.
Orr RK. 1995. Use of probabilistic neural networks to predict mortality following cardiac surgery. Proceedings of WCNN. Vol II. p. 754–757.
Davalo E, Niam P. 1990. Neural Networks. Macmillan Press, p. 111–112.
Tan H, Prokhorov DV and Wunsch DC. 1995. Probabilistic and time-delay neural network techniques for conservative short-term stock trend prediction. Proceedings of World Congress on Neural Networks. Vol II. p. 44–47.
Fu L. 1994. Neural Networks in Computer Intelligence. McGraw-Hill International Editions, p.351–369.
Andrews R, Diederich J and Tickle AB. 1995. A survey and critique of techniques for extracting rules from trained Artificial Neural Networks. Knowledge Based Systems Vol 8 (6, December) p. 373–389.
Sestito S and Dillon T. 1994. Automatic knowledge acquisition. Prentice Hall.
Towell G and Shavlik J. 1993. The extraction of refined rules from knowledge based neural networks. Machine Learning Vol 131, p71–101.
Karin ED. 1990. A simple procedure for pruning back-propagation trained neural networks. I.E. Trans. Neural Networks, vol.1 no.2. p239–242.
Weigend AS, Rumelhart DE and Huberman BA. 1991. Generalization by weight elimination with applications to forecasting. In Advances in Neural Information Processing (3). Lippmann R, Moody J and Touretzky D. Eds. p. 875–882.
Balls GR, Palmer-Brown D, Cobb AH and Sanders GE. 1995. Towards unravelling the complex interactions between microclimate, ozone dose and ozone injury in clover. Journal of Water, Air and Soil Pollution. 85, 1467–1472.
Balls GR, Palmer-Brown D, and Sanders GE. 1996. Investigating microclimate influences on ozone injury in clover (Trifolium subterraneum) using artificial neural networks. New Phytologist, 132, 271–280
Roadknight CM, Palmer-Brown D and Sanders GE. 1995. Learning the equations of data. Proceedings of 3rd annual SNN symposium on neural networks (eds. Kappen B and Gielen S) Springer-Verlag. 253–257.
Roadknight CM, Balls GR, Sanders GE and Palmer-Brown D. Modelling complex environmental data. IEEE Transactions on Neural Networks: Special edition on everyday applications. (In Press July 1997.)
Sietsma J and Dow RJF. 1988. Neural net pruning — Why and how. Prc. IEEE Int. Conf. Neural Networks. Vol 1. p. 325–333.
Wiersma FR, Poel M and Oudshoff AM. 1995. The BB neural network rule extraction method. Proceedings of 3rd annual SNN symposium on neural networks (eds. Kappen B and Gielen S) Springer-Verlag. 69–73.
Ripley BD. 1995. Statistical ideas for selecting network architectures. Proceedings of 3rd annual SNN symposium on neural networks (eds. Kappen B and Gielen S) Springer-Verlag. 183–190.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1997 Springer-Verlag
About this paper
Cite this paper
Roadknight, C.M., Palmer-Brown, D., Mills, G.E. (1997). The analysis of artificial neural network data models. In: Liu, X., Cohen, P., Berthold, M. (eds) Advances in Intelligent Data Analysis Reasoning about Data. IDA 1997. Lecture Notes in Computer Science, vol 1280. Springer, Berlin, Heidelberg. https://doi.org/10.1007/BFb0052852
Download citation
DOI: https://doi.org/10.1007/BFb0052852
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-63346-4
Online ISBN: 978-3-540-69520-2
eBook Packages: Springer Book Archive