Combined Neural Network Models for Epidemiological Data: Modelling Heterogeneity and Reduction of Input Correlations
We consider an epidemiological dataset, concerned with predicting pulmonary response to air pollution. To gain more knowledge of nonlinear effects and interactions in the data, nonlinear neural network techniques were applied to model the data. Initially, we modelled the data with standard feedforward network models. Based on the epidemiologic effect of heterogeneity in response, we propose a novel combined neural network modelling strategy to improve prediction quality. Also, we propose the use of a neural network strategy for reducing correlation between covariants to improve modelling quality. The results presented are promising when compared to standard feedforward network modelling.
KeywordsInput Pattern Peak Expiratory Flow Feedforward Network Learn Vector Quantization Pulmonary Response
Unable to display preview. Download preview PDF.
- D. DeMers and G.W. Cottrell. Nonlinear dimensionality reduction. NIPS 5, pp.580–587, 1993.Google Scholar
- J. Hertz, A. Krogh, and R.G. Palmer. Introduction to the Theory of Neural Computation, volume 1 of Santa Fe Institute Studies in the Sciences of Complexity. Addison Wesley, 1991.Google Scholar
- T. Kohonen. The self-organizing map. Proceedings of the IEEE, 78(9):1464–1480, 1990.Google Scholar
- B.D. Ripley. Statistical aspects of neural networks. In Séminaire Européen de Statistique, 1992.Google Scholar
- D.A. Savitz. In defense of black box epidemiology. Epidemiology, 5:550–552, 1994.Google Scholar
- P. Skrabanek. The emptiness of the black box. Epidemiology, 5:553–555, 1994.Google Scholar
- T. Waschulzik et al. Evaluation of an epidemiological data set as an example of the application of neural networks to the analysis of large medical data sets. 4 th Conf AI in Medicine Europe, pp.466–476, 1993.Google Scholar