Abstract
One of the main limitations of artificial neural networks (ANN) is their high inability to know in an explicit way the relations established between explanatory variables (input) and dependent variables (output). This is a major reason why they are usually called “black boxes.” In the last few years, several methods have been proposed to assess the relative importance of each explanatory variable. Nevertheless, it has not been possible to reach a consensus on which is the best-performing method. This is largely due to the different relative importance obtained for each variable depending on the method used. This importance also varies with the designed network architecture and/or with the initial random weights used to train the ANN. This paper proposes a procedure that seeks to minimize these problems and provides consistency in the results obtained from different methods. Essentially, the idea is to work with a set of neural networks instead of a single one. The proposed procedure is validated using a database collected from a customer satisfaction survey, which was conducted on the public transport system of Granada (Spain) in 2007. The results show that, when each method is applied independently, the variable’s importance rankings are similar and, in addition, coincide with the hierarchy established by researchers who have applied other techniques.
Similar content being viewed by others
References
Akin D, Akbaç B (2010) A neural network (NN) model to predict intersection crashes based upon driver, vehicle and roadway surface characteristics. Sci Res Essays 5(19):2837–2847
Azadeh A, Rouzbahman M, Saberi M, Fam IM (2011) An adaptative neural network algorithm for assessment and improvement of job satisfaction with respect to HSE and ergonomics program: the case of a gas refinery. J Loss Prev Process Ind 24:361–370
Beale MH, Hagan MT, Demuth HB (2007) Neural Network Toolbox 7. User’s Guide. MathWorks, Inc. 3 Apple Hill Drive Natic, MA 01760-2098
Cao M, Qiao P (2008) Neural network committee-based sensitivity analysis strategy for geotechnical engineering problems. Neural Comput Appl 17:509–519
Cortez P, Embrechts MJ (2011) Opening black box data mining models using sensitivity analysis. IEEE symposium series in computational intelligence, Paris, France, 4, 2011
Cortez P, Embrechts MJ (2013) Using sensitivity analysis and visualization techniques to open black box data mining models. Inf Sci 225:1–17
De Oña J, De Oña R, Calvo FJ (2012) A classification tree approach to identify key factors of transit service quality. Expert Syst Appl 39:11164–11171
De Oña J, De Oña R, Eboli L, Mazzulla G (2013) Perceived service quality in bus transit service: a structural equation approach. Transp Policy 29:219–226
Delen D, Sharda R, Bessonov M (2006) Identifying significant predictors of injury severity in traffic accidents using a series of artificial neural networks. Accid Anal Prev 38:434–444
Dell’Olio L, Ibeas A, Cecín P (2010) Modelling user perception of bus transit quality. Transp Policy 17(6):388–397
Dell’Olio L, Ibeas A, Cecín P (2011) The quality of service desired by public transport users. Transp Policy 18(1):217–227
Dimopoulos Y, Bourret P, Lek S (1995) Use of some sensitivity criteria for choosing networks with good generalization ability. Neural Process Lett 2:1–4
Eboli L, Mazulla G (2008) Willingness-to-pay of public transport users for improvement in service quality. Eur Transp 38:107–118
Eboli L, Mazulla G (2010) How to capture the passengers’ point of view on a transit service through rating and choice opinions. Transp Rev 30:435–450
Engelbrecht AP, Cloete I, Zurada JM (1995) Determining the significance of input parameters using sensitivity analysis, from natural to artificial neural computation. In: proceedings of International Workshop on Artificial Neural Networks. Málaga-Torremolinos, Springer, Spain, pp 382–388
Funahashi KI (1989) On the approximate realization of continuous mappings by neural networks. Neural Netw 2:183–192
Garson GD (1991) Interpreting neural-network connection weights. Artif Intell Expert 6:47–51
Gedeon, T.D., Wong, P.M. & Harris, D., (1995). Balancing the bias and variance: network topology and pattern set reduction techniques. In: proceedings of International Workshop on Artificial Neural Networks, IWANN95, Torremolinos, España, pp 550–558
Gedeon TD (1997) Data mining of inputs: analyzing magnitude of functional measures. Int J Neural Syst 8(2):209–218
Gevrey M, Dimopoulos I, Lek S (2003) Review and comparison of methods to study the contribution of variables in artificial neural network models. Ecol Model 160:249–264
Gevrey M, Dimopoulos I, Lek S (2006) Two-way interaction of input variables in the sensitivity analysis of neural network models. Ecol Model 195:43–50
Goh ATC (1995) Back-propagation neural networks for modeling complex systems. Artif Intell Eng 9:143–151
Hagan MT, Demuth HB, Beale MH (1996) Neural network design. Campus Publishing Service, Colorado University Bookstore, Colorado. ISBN 0-9717321-0-8
He F, Sung AH, Guo B (1997) A neural network for prediction of oil well cement bonding quality. In: proceedings of IASTED international conference on control, IASTED-ACTA Press, Cancun-Mexico, pp 417–420
Hunter A, Kennedy L, Henry J, Ferguson I (2000) Application of neural networks and sensitivity analysis to improved prediction of trauma survival. Comput Methods Progr Biomed 62:11–19
Hornik K, Stichcombe M, White H (1989) Multilayer feedforward networks are universal approximators. Neural Netw 2:359–366
Kemp SL, Zaradic P, Hansen F (2007) An approach for determining relative input parameter importance and significance in artificial neural networks. Ecol Model 204:326–334
Kewley R, Embrechts M, Breneman C (2000) Data strip mining for the virtual design of pharmaceuticals with neural networks. IEEE Trans Neural Netw 11(3):668–679
Lek S, Beland A, Dimopoulos I, Lauga J, Moreau J (1995) Improved estimation, using neural networks, of the food consumption of fish populations. Mar Freshw Res 46(8):1229–1236
Lek S, Delacoste M, Baran P, Dimopoulos I, Lauga J, Aulagnier S (1996) Application of neural networks to modeling nonlinear relationships in ecology. Ecol Model 90:39–52
Lek S, Beland A, Baran P, Dimopoulos I, Delacoste M (1996) Role of some environmental variables in trout abundance models using neural networks. Aquat Living Resour 9:23–29
Lin Y, Cunningham GA (1995) A new approach to fuzzy-neural system modeling. IEEE Trans Fuzzy Syst 3(2):190–198
Martín del Bío B, Sanz Molina A (2006) Neural networks and fuzzy systems. Editorial RA-MA
Masters T (1993) Practical neural networks recipes in C ++. Academic Press, Waltham
Moghaddam FR, Afandizadeh S, Ziyadi M (2010) Prediction of accident severity using artificial neural networks. Int J Civ Eng 9:1
Mohammadipour AH, Alavi SH (2009) The optimization of the geometric cross-section dimensions of raised pedestrian crosswalks: a case of study in Qazvin. Accid Anal Prev 41:314–326
Mussone L, Ferrari A, Oneta M (1999) An analysis of urban collisions using an artificial intelligence model. Accid Anal Prev 31:705–718
Olden JD, Jackson DA (2002) Illuminating the “black-box”: a randomization approach for understanding variable contributions in artificial neural networks. Ecol Model 154:135–150
Olden JD, Joy MK, Death RG (2004) An accurate comparison of methods for quantifying variable importance in artificial neural networks using simulated data. Ecol Model 178:389–397
Özesmi SL, Özesmi U (1999) An artificial neural network approach to spatial habitat modeling with interspecific interaction. Ecol Model 116:15–31
Paliwal M, Kumar UA (2011) Assessing the contribution of variables in feed forward neural network. Appl Soft Comput 11:3690–3696
Palmer A, Montaño JJ (2002a) Redes neuronales artificiales aplicadas al análisis de datos. Doctoral Dissertation. University of Palma de Mallorca
Palmer A, Montaño JJ (2002) Redes neuronales artificiales: abriendo la caja negra. Metodología de las ciencias del comportamiento 4(1):77–93
Palmer A, Montaño JJ (2003) Numeric sensitivity analysis applied to feed forward neural networks. Neural Comput Appl 12:119–125
Rzempoluk EJ (1998) Neural network data analysis using simulnet. Springer, New York
Rumelhart DE, McClelland JL (1986) Parallel distributed processing, vol 1: foundations. MIT Press, Cambridge
Rumelhart DE, Hinton GE, Williams RJ (1986) Learning representations by backpropagation errors. Nature 323:533–536
Scardi M, Harding LW (1999) Developing an empirical model of phytoplankton primary production: a neural networks case study. Ecol Model 120(2–3):213–223
Sung AH (1998) Ranking importance of input parameters of neural networks. Expert Syst Appl 15:405–411
Watts MJ, Worner SP (2008) Using artificial neural networks to determine the relative contribution of abiotic factors influencing the establishment of insect pest species. Ecol Inform 3:64–74
Werbos PJ (1974) Beyond regression: new tools for prediction and analysis in behavioral sciences. Doctoral Dissertation. Applied Mathematics, Harvard University
Yao J, Teng N, Poh HL, Tan CL (1998) Forecasting and analysis of marketing data using neural networks. J Inf Sci Eng 14:843–862
Yeh I, Cheng W (2010) First and second order sensitivity analysis of MLP. Neurocomputing 73:2225–2233
Zhou ZH, Wu J, Tang W (2002) Ensembling neural networks: many could be better than all. Artif Intell 137(1–2):239–263
Zurada JM, Malinowski A, Cloete I (1994) Sensitivity analysis for minimization of input data dimensión for feed forward neural network. In: proceedings of IEEE international symposium on circuits and systems, IEEE Press, London
Acknowledgments
Support from Consejería de Innovación, Ciencia y Economía of the Junta de Andalucía (Spain) (Research Project P08-TEP-03819, co-funded by FEDER) is gratefully acknowledged. The authors also acknowledge the Granada Consorcio de Transportes for making the data set available for this study.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
de Oña, J., Garrido, C. Extracting the contribution of independent variables in neural network models: a new approach to handle instability. Neural Comput & Applic 25, 859–869 (2014). https://doi.org/10.1007/s00521-014-1573-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-014-1573-5