Prediction of dilution in sublevel stoping through machine learning algorithms

Abstract

Abstract One of the most used underground mining methods is open stope mining which involves extracting a large body of ore through drilling and blasting. The method offers plenty of advantages but it has some very important drawbacks, such as overbreak, wall instability and unplanned ore dilution. The research looks to test the efficiency of using machine learning algorithms to estimate the dilution in open stopes, some of the expected benefits are reduced time cost (compared to numerical analysis) and more accurate results (compared to empirical methods). The algorithms utilized are: random forest (RF), support vector machine (SVM) and k-nearest neighbors (KNN). The algorithms were trained and tested with 752 cases from several mines across the world. Three algorithms accomplished AUC scores over 0.850, which can be considered excellent results, but random forest achieved the most impressive results (precision score = 0.835, accuracy score = 0.804 AUC score = 0.942). From the obtained results it is possible to conclude that the machine learning algorithms can be used as trustworthy tools for the estimation of dilution, but some adjustments may be needed to increase the accuracy to specific mine sites.