Machine learning algorithms for rock cutting slopes stability condition identification

Abstract

Transportation systems play a fundamental role in nowadays society. Indeed, every developed or countries undergoing development have invested and keep investing to build a safe and functional transportation network. The main concern nowadays, particularly for developed countries that already have a very complete network, is to keep it operational under all conditions. However, due to the network extension and increased budget constraints, such task is difficult to accomplish. In the framework of transportations networks, particularly for railway, slopes are perhaps the element for which their failure can have a strongest impact at several levels. Although there are some models and systems to detect slope failures, most of them were developed for natural slopes, presenting some constrains when applied to engineered (human-made) slopes. They have limited applicability as most of the existing systems were developed based on particular case studies or using small databases. Moreover, another aspect that can limit its applicability is related with the information used to feed them, such as data taken from complex tests or from expensive monitoring systems. Aiming to overcome this drawback, we took advantage of the high flexible learning capabilities of Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs), which have been used in the past to model complex nonlinear mappings. Both data mining algorithms were applied in the development of a classification tool able to identify the stability condition of a rock cutting slope, keeping in mind the use of information usually collected during routine inspections activities (visual information) to feed them. For that, two different strategies were followed: nominal classification and regression. Moreover, to overcome the problem of imbalanced data, three training sampling approaches were explored: no resampling, SMOTE and Oversampling. The achieved results are present and discussed, comparing the performance of both algorithms (ANN and SVM) according to each modeling strategy as well as the effect of the sampling approaches.