Investigation into the robustness of artificial neural networks for a case study in civil engineering
Mohamed A. Shahin, Holger Robert Maier, Mark B. Jaksa · Adelaide Research & Scholarship (AR&S) (University of Adelaide) · 2005
Artificial neural networks (ANNs) have been used as a prediction tool in many areas of engineering.In order to test the robustness and generalisation ability of ANN models, the approach that is generally adopted is to test the performance of trained ANNs on an independent validation set.If such performance is adequate, the model is deemed to be robust and able to generalise.However, this is not necessarily the case.In this paper, the robustness of ANN models is investigated for a case study of predicting the settlement of shallow foundations on granular soils.A procedure that tests the robustness of the predictive ability of ANN models is introduced.The results indicate that good performance of ANN models on the data used for model calibration and validation does not guarantee that the models will perform in a robust fashion over a range of data similar to that used in the model calibration phase.The results also indicate that validating ANN models using the procedure provided in this study is essential in order to investigate their robustness. 1