Construction Engineering Decision Support Enabled by Minimization of Localized Generalization Error of RBFNN

In an attempt to address the limitations of neural modeling in engineering applications, we draw on a novel localized generalization error model to train RBFNN on a dataset obtained from the domain of construction engineering and project management, with the objective of devising reliable strategies that may shorten the cycle time required for constructing one span of precast viaduct. We select the RBFNN with the optimal number of hidden neurons, which yields the maximal coverage around training samples given a predetermined generalization error bound. By analyzing the values of center vectors of RBFNN, we take one step further to uncover hidden patterns or rules by which RBFNN maps input features onto output classifications. The rules derived from the model are well corroborated by domain experts and computer simulation models