Construction of a predictive model of shovel productivity applying machine learning algorithms

Currently, many mining companies face the need to increase the efficiency of their performance in conditions of growing competitiveness and market globalization. To achieve the set objectives, it is essential to reduce costs, optimize the use of equipment and human resources. Mining companies usually implement technological process control systems where a large amount of updated information about their production processes is stored, normally using long-term accumulated information that consists mainly of statistical reports. Taking the above into account, the need arises to use this accumulated information to identify segments of technological processes that are less efficient and be able to improve productivity. To do this, new technologies can be used such as the use of Machine Learning (ML). The present work focuses on the creation of a predictive model of “shovel” productivity, where shovels are heavy load equipment, used for any activity that involves moving earth or rocks in large volumes, with a level of acceptable confidence, which unlike the traditional method, include other quantitative and qualitative variables. The research began with the general hypothesis that the productivity model developed could improve the prediction of blade productivity with greater accuracy using Machine Learning models. From the results we obtained that the model that best predicts the productivity of the blades is the “Random Forest Regressor” with an accuracy of 99.7% for the training and 91.1% for the testing DataSet; while the worst Machine Learning precision model was “Dummy Regressor”.