Designing neural networks for modeling biological data: A statistical perspective
Open Access
- 1 January 2014
- journal article
- Published by American Institute of Mathematical Sciences (AIMS) in Mathematical Biosciences and Engineering
- Vol. 11 (2) , 331-342
Abstract
In this paper, we propose a strategy for the selection of the hidden layer size in feedforward neural network models. The procedure herein presented is based on comparison of different models in terms of their out of sample predictive ability, for a specified loss function. To overcome the problem of data snooping, we extend the scheme based on the use of the reality check with modifications apt to compare nested models. Some applications of the proposed procedure to simulated and real data sets show that it allows to select parsimonious neural network models with the highest predictive accuracy.Keywords
This publication has 34 references indexed in Scilit:
- In-sample tests of predictive ability: A new approachJournal of Econometrics, 2012
- Reality Checks and Comparisons of Nested Predictive ModelsJournal of Business & Economic Statistics, 2012
- Artificial Neural Networks in Biology and Chemistry—The Evolution of a New Analytical Tool2008
- Predictive Density EvaluationSSRN Electronic Journal, 2005
- Optimal forecast combinations under general loss functions and forecast error distributionsJournal of Econometrics, 2004
- On the Use of Neural Network Ensembles in QSAR and QSPRJournal of Chemical Information and Computer Sciences, 2002
- Neural network schemes for detecting rare events in human genomic DNABioinformatics, 2000
- Improved rates and asymptotic normality for nonparametric neural network estimatorsIEEE Transactions on Information Theory, 1999
- Prediction Intervals for Neural Networks via Nonlinear RegressionTechnometrics, 1998
- Universal approximation bounds for superpositions of a sigmoidal functionIEEE Transactions on Information Theory, 1993