Forecasting mining capital cost for open-pit mining projects based on artificial neural network approach

Abstract

This study considered and developed four artificial intelligence (AI) techniques to estimate mining capital cost (MCC) for open-pit copper mining projects with high accuracy, including artificial neural network (ANN), random forest (RF), support vector machine (SVM), and classification and regression tree (CART); 74 observations of mining projects were collected and analyzed to predict MCC based on five input variables. Root-mean-squared error (RMSE), coefficient of correlation (R2), mean absolute error (MAE), and absolute percentage error (APE), were used to evaluate the performance/quality/accuracy of the models. The results of this study indicated that ANN, RF, SVM and CART models were advanced techniques in predicting MCC with high accuracy. Of those, the ANN model yielded the most dominant accuracy/performance with an RMSE of 138.103, R2 of 0.990, MAE of 114.589, and APE of 7.770%. The remaining models (i.e. RF, SVM, CART) yielded lower performance with RMSE in the range of 172.975–379.691, R2 in the range of 0.924–0.987, MAE in the range of 134.982–301.196, and APE in the range of 10.339%–19.384%. The results of the sensitivity analysis of this work also revealed that production capacity includes MineAP and MillAP, were the two most essential parameters on the MCC predictive models. They should be used as the primary input parameters for estimating MCC in actual.