Global Sensitivity Estimates for Neural Network Classifiers

Artificial neural networks (ANNs) have traditionally been seen as black-box models, because, although they are able to find “hidden” relations between inputs and outputs with a high approximation capacity, their structure seldom provides any insights on the structure of the functions being approximated. Several research papers have tried to debunk the black-box nature of ANNs, since it limits the potential use of ANNs in many research areas. This paper is framed in this context and proposes a methodology to determine the individual and collective effects of the input variables on the outputs for classification problems based on the ANOVA-functional decomposition. The method is applied after the training phase of the ANN and allows researchers to rank the input variables according to their importance in the variance of the ANN output. The computation of the sensitivity indices for product unit neural networks is straightforward as those indices can be calculated analytically by evaluating the integrals in the ANOVA decomposition. Unfortunately, the sensitivity indices associated with ANNs based on sigmoidal basis functions or radial basis functions cannot be calculated analytically. In this paper, the indices for those kinds of ANNs are proposed to be estimated by the (quasi-) Monte Carlo method.

[1]  César Hervás-Martínez,et al.  Multinomial logistic regression and product unit neural network models: Application of a new hybrid methodology for solving a classification problem in the livestock sector , 2009, Expert Syst. Appl..

[2]  Hongming Zhou,et al.  Extreme Learning Machine for Regression and Multiclass Classification , 2012, IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics).

[3]  Ignacio Requena,et al.  Are artificial neural networks black boxes? , 1997, IEEE Trans. Neural Networks.

[4]  Andries P. Engelbrecht,et al.  A sensitivity analysis algorithm for pruning feedforward neural networks , 1996, Proceedings of International Conference on Neural Networks (ICNN'96).

[5]  Paulo J. G. Lisboa,et al.  The Interpretation Of Supervised Neural Networks , 1993, Workshop on Neural Network Applications and Tools.

[6]  Flavio Cannavó,et al.  Sensitivity analysis for volcanic source modeling quality assessment and model selection , 2012, Comput. Geosci..

[7]  César Hervás-Martínez,et al.  Addressing the EU Sovereign Ratings Using an Ordinal Regression Approach , 2013, IEEE Transactions on Cybernetics.

[8]  Minoru Fukumi,et al.  A new rule extraction method from neural networks , 1999, IJCNN'99. International Joint Conference on Neural Networks. Proceedings (Cat. No.99CH36339).

[9]  Harvey M. Wagner,et al.  Global Sensitivity Analysis , 1995, Oper. Res..

[10]  Sheng Chen,et al.  Orthogonal least squares methods and their application to non-linear system identification , 1989 .

[11]  Jacek M. Zurada,et al.  Perturbation method for deleting redundant inputs of perceptron networks , 1997, Neurocomputing.

[12]  Yannis Dimopoulos,et al.  Use of some sensitivity criteria for choosing networks with good generalization ability , 1995, Neural Processing Letters.

[13]  F. J. Martı́nez-Estudilloa,et al.  Evolutionary product-unit neural networks classifiers , 2008 .

[14]  Robert H. Kewley,et al.  Data strip mining for the virtual design of pharmaceuticals with neural networks , 2000, IEEE Trans. Neural Networks Learn. Syst..

[15]  Jonathan D. Buckley,et al.  Predicting Time-to-Relapse in Breast Cancer Using Neural Networks , 1997 .

[16]  Joachim Diederich,et al.  Survey and critique of techniques for extracting rules from trained artificial neural networks , 1995, Knowl. Based Syst..

[17]  Saltelli Andrea,et al.  Global Sensitivity Analysis: The Primer , 2008 .

[18]  I. Sobol,et al.  On quasi-Monte Carlo integrations , 1998 .

[19]  Robert H. Kewley,et al.  Data Mining for Molecules with 2-D Neural Network Sensitivity Analysis , 2003 .

[20]  Jacek M. Zurada,et al.  Extracting Rules From Neural Networks as Decision Diagrams , 2011, IEEE Transactions on Neural Networks.

[21]  D. Sargent,et al.  Comparison of artificial neural networks with other statistical approaches , 2001, Cancer.

[22]  César Hervás-Martínez,et al.  JCLEC: a Java framework for evolutionary computation , 2007, Soft Comput..

[23]  Mian Li,et al.  Robust Optimization and Sensitivity Analysis with Multi-Objective Genetic Algorithms: Single- and Multi-Disciplinary Applications , 2007 .

[24]  Andries P. Engelbrecht,et al.  Pruning product unit neural networks , 2002, Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN'02 (Cat. No.02CH37290).

[25]  Kezhi Mao,et al.  Fast orthogonal forward selection algorithm for feature subset selection , 2002, IEEE Trans. Neural Networks.

[26]  A M Walker,et al.  Epidemiologic interpretation of artificial neural networks. , 1998, American journal of epidemiology.

[27]  I. Sobol.  Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates , 2001 .

[28]  Andries Petrus Engelbrecht,et al.  A new pruning heuristic based on variance analysis of sensitivity information , 2001, IEEE Trans. Neural Networks.

[29]  Bertrand Iooss,et al.  Global sensitivity analysis of stochastic computer models with joint metamodels , 2008, Statistics and Computing.

[30]  Daniel S. Yeung,et al.  Sensitivity analysis of multilayer perceptron to input and weight perturbations , 2001, IEEE Trans. Neural Networks.

[31]  Russell G. Death,et al.  An accurate comparison of methods for quantifying variable importance in artificial neural networks using simulated data , 2004 .

[32]  C. Fortuin,et al.  Study of the sensitivity of coupled reaction systems to uncertainties in rate coefficients. I Theory , 1973 .

[33]  Olivier Roustant,et al.  Calculations of Sobol indices for the Gaussian process metamodel , 2008, Reliab. Eng. Syst. Saf..

[34]  A. Saltelli,et al.  A quantitative model-independent method for global sensitivity analysis of model output , 1999 .

[35]  Eric Fock,et al.  Global Sensitivity Analysis Approach for Input Selection and System Identification Purposes—A New Framework for Feedforward Neural Networks , 2014, IEEE Transactions on Neural Networks and Learning Systems.

[36]  Philip D. Wasserman,et al.  Advanced methods in neural computing , 1993, VNR computer library.

[37]  Bernard Widrow,et al.  Sensitivity of feedforward neural networks to weight errors , 1990, IEEE Trans. Neural Networks.

[38]  David E. Rumelhart,et al.  Product Units: A Computationally Powerful and Biologically Plausible Extension to Backpropagation Networks , 1989, Neural Computation.

[39]  Sundaram Suresh,et al.  Meta-cognitive RBF Network and its Projection Based Learning algorithm for classification problems , 2013, Appl. Soft Comput..

[40]  Julian D. Olden,et al.  Illuminating the “black box”: a randomization approach for understanding variable contributions in artificial neural networks , 2002 .

[41]  John H. Seinfeld,et al.  Global sensitivity analysis—a computational implementation of the Fourier Amplitude Sensitivity Test (FAST) , 1982 .

[42]  Lucila Ohno-Machado,et al.  Logistic regression and artificial neural network classification models: a methodology review , 2002, J. Biomed. Informatics.

[43]  S. Hashem,et al.  Sensitivity analysis for feedforward artificial neural networks with differentiable activation functions , 1992, [Proceedings 1992] IJCNN International Joint Conference on Neural Networks.

[44]  César Hervás-Martínez,et al.  Determination of relative agrarian technical efficiency by a dynamic over-sampling procedure guided by minimum sensitivity , 2011, Expert Syst. Appl..

[45]  Pedro Antonio Gutiérrez,et al.  Evolutionary product-unit neural networks classifiers , 2008, Neurocomputing.

[46]  D. M. Titterington,et al.  Neural Networks: A Review from a Statistical Perspective , 1994 .

[47]  Holger R. Maier,et al.  A comparison of sensitivity analysis techniques for complex models for environmental management , 2005 .

[48]  Annalisa Riccardi,et al.  Ordinal Regression by a Generalized Force-Based Model , 2015, IEEE Transactions on Cybernetics.

[49]  Tharam S. Dillon,et al.  Knowledge acquisition of conjunctive rules using multilayered neural networks , 1993, Int. J. Intell. Syst..

[50]  G. David Garson,et al.  Interpreting neural-network connection weights , 1991 .

[51]  Pedro Antonio Gutiérrez,et al.  Hybridizing logistic regression with product unit and RBF networks for accurate detection and prediction of banking crises , 2010, Omega.

[52]  Andries Petrus Engelbrecht,et al.  Incremental learning using sensitivity analysis , 1999, IJCNN'99. International Joint Conference on Neural Networks. Proceedings (Cat. No.99CH36339).