Equipment downtime, whether planned or unplanned, results in severe penalties and significant losses as well as the degradation and termination of the primary operation in manufacturing machines. In this paper, we explore predictive methods and develop models for predictive maintenance that use machine learning algorithms to anticipate downtime on manufacturing lines. Typically, computer numerical control (CNC) machines generate a large number of data points that can be used by manufacturers to predict downtimes. However, for operators in low-resource and constrained environments, such data might be difficult to obtain and analyse. The main goals of the study are to determine if data on current drawn and machine cycles can help in predicting downtimes, and which low-resource machine learning model performs best. In particular, we report on experiments that compare the performance of random forest and XGBoost to determine which model can offer superior performance. Our results indicate random forest outperforming XGBoost in terms of accuracy, recall, and precision.