关键词: 神经网络, 模型选择, 统计检验

Abstract: How statistical and mathematical tools commonly used independently can advantageously be exploited together was studied in order to improve neural network estimation and selection in nonlinear static modeling. The statistical tools considered were the analysis of the numerical conditioning of the neural network candidates, statistical hypothesis tests, and cross validation. Each of these tools was presented and analyzed in order to justify at what stage of a construction and selection procedure they can be most useful. On the basis of this analysis, a novel and systematic construction and selection procedure for neural modeling was proposed. In the end, its efficiency was illustrated through simulations.

Key words: neural networks, model selection, statistical hypothesis tests