Use of machine learning techniques to analyse the risk associated with mine sludge deposits
Mathematical and Computer Modelling: An International Journal
| Hi-index | 0.00 |
Given the non-existence of a standard quality index for ornamental slate, granite and marble blocks, classical classification methods such as Bieniawski's rock mass rating (RMR) and Barton's Q System are typically used, very much unquestioningly and indiscriminately. These methods, however, were originally designed for a different purpose, namely to identify and solve stability problems in underground excavations. Furthermore, they rely on pre-established parametric formulas that do not necessarily coincide in terms of criteria with rock quality experts. In fact, the fit between both perspectives has never been properly evaluated. In this article, we propose a quality index for ornamental rock constructed using machine learning techniques (support vector machines (SVMs)) to model the quality grade allocation procedure as applied by the expert. This quality index is composed of a set of logical rules consisting of mathematical inequalities in relevant ornamental quality variables. Nonetheless, the fact that the SVMs have the drawback of being difficult to interpret complicates the process of extracting the quality rules used by an expert. To overcome this drawback, we used classification and regression trees (CARTs) trained using SVM output.