This study developed machine learning (ML) models for predicting rock formation reactive forces experienced by a hydraulic shovel bucket during excavation. To do this, rock formation in the form of muckpile was modeled as granular ball elements in three-dimensional particle flow code (PFC) using linear bonding logic. A typical field-size hydraulic shovel bucket modelled in AutoCAD was used to excavate the rock formation and the three-dimensional shovel bucket resistive forces required to cut through the rock formation were measured and recorded. The experiments involved different muckpile height, bulk densities, repose angles, and average fragment size. Six ML algorithms, artificial neural network (ANN), K-nearest neighbor (KNN), linear, random forest (RF), regression tree (RT), and support vector machine (SVM), were evaluated on their ability to predict the shovel bucket resistive forces. The input variables used for the prediction of the resistive forces were the muckpile bulk density, angle of repose, height, and fragment size. The results showed that the ML techniques are useful and reliable methods of predicting rock formation resistive forces based on the rock formation properties. This research is a preliminary step towards developing reliable models for accurate prediction of excavation forces/torques which could potentially enhance hydraulic shovel process automation.