Subsampling Model Selection in Neural Networks for Nonlinear Time Series Analysis
Michele La Rocca · 2004
In this paper, the subsampling method is applied to the problem of model selection in neural networks for non linear time series data. A complete strategy, which combines a set of graphical, exploratory and inferential statistical tools, is proposed to select the topology of a neural network model. The procedure allows to choose the number and the type of inputs (by using a formal test procedure based on relevance measures) and to identify the hidden layer size (by looking at the predictive performance of the neural network model). The proposed approach heavily uses the subsampling technique to extend some approaches already available for the iid case to dependent data. 1