Abstract:

While building a resource model, the interpretations made by geologists about the continuity of geological domains give rise to uncertainties. Quantifying these uncertainties can be a significant advantage for the evaluation of mineral resources, however, it can be a challenging process. While some traditional geostatistical tools such as simulations are viable options to quantify the uncertainty, the mining industry can incorporate other technologies for such challenges. In this context, machine learning offers promising solutions to take advantage of the data available and permits the quantification of uncertainties associated with these models. The number of applications involving machine learning algorithms keeps growing every day. Although machine learning algorithms have been applied in several studies successfully, bias and variance are problems to deal with while building a model. Ensemble learning, which is the technique combining multiple models to build an accurate model, is a robust solution with an optimum bias and variance. In this study, we start with an ensemble implementation of an unsupervised - binary clustering method, which takes geochemical data and splits it into the two most distinct subsets. These labeled data then inform a support vector classification algorithm, which is a supervised machine learning method. The proposed workflow is repeated by building classifiers with subsets of training samples (bagging) and subsets of variables, which leads to a collection of weak classifiers that together define a stronger classifier. The methodology is applied to real geochemical data, and Sequential Indicator Simulation is performed to compare and validate the model. The method can be applied hierarchically to define multiple domains.

Menu

Mr. KASIMCAN KORUK

Biography