Artificial neural networks for rock and soil cutting slopes stability condition prediction

Abstract

This study aims to develop a tool able to help decision makers to find the best strategy for slopes management tasks. It is known that one of the main challenges nowadays for every developed or countries undergoing development is to keep operational under all conditions their transportations infrastructure. However, due to the network extension and increased budget constraints such challenge is even more difficult to accomplish. Keeping in mind the strong impact of a slope failure in the transportation infrastructure it is important to develop tools able to help minimizing this situation. Accordingly, and in order to achieve this goal, the high flexible learning capabilities of Artificial Neural Networks (ANNs) were applied in the development of a classification tool aiming to identify the stability condition of a rock and soil cutting slopes, keeping in mind the use of information usually collected during routine inspections activities (visual information) to feed them. For that, it was followed a nominal classification strategy and, in order to overcome the problem of imbalanced data, three training sampling approaches were explored: no resampling, SMOTE (Synthetic Minority Over-sampling Technique) and Oversampling. The achieved results are presented and discussed, comparing the achieved performance for both slope types (rock and soil cuttings) as well as the effect of the sampling approaches. An input-sensitivity analysis was applied, allowing to measure the relative influence of each model attribute.