ABSTRACT
This article describes the construction of an expert system to evaluate the quality of a slate quarry. Different machine learning techniques—classification trees, support vector machines and Bayesian networks -were used with a view to evaluating and comparing interpretability, prediction capacity, and facility for incorporating a priori information in the model. The three techniques contribute in complementary ways as a result of their different internal configurations and characters (discriminative or generative). The Bayesian networks produced the most satisfactory results for our slate problem, given that they combine both predictive and descriptive capacities.
