Using machine learning algorithms to predict cast blasting performance in surface mining

ABSTRACT Overburden removal is a major activity of surface coal mining and accounts for over 60–70% of the costs. Cast blasting is integral to overburden removal using draglines. Knowledge of cast blasting was combined with data analytics and machine learning algorithms to predict cast blast percentage. In a typical study, the cast percentage is predicted as a function of key input variables, namely (1) height to burden (H/b) ratio, (2) height to width (H/W) ratio, (3) length to width (L/W) ratio, (4) effective in-hole explosive density (de – te/m3), (5) powder factor (PF) (m3/kg – volume of rock broken per kg of explosive), and (6) average delay per unit width of burden (ms/m). Random forest algorithm was used under five-fold cross-validation with 68 datasets split into 57 for training and 11 for testing purposes. The model produced an R 2 value of 69.16% and 67.37% respectively on the training and testing data.