The objective of this study is to identify the most effective machine learning algorithm for predictive maintenance of industrial machinery using three input variables: temperature, vibration, and machine condition. Considering the balance between predictive accuracy and computational efficiency, as well as the practicality of implementation in resource-constrained environments. This study evaluated the effectiveness of six machine learning algorithms for predictive maintenance in industrial environments using three input variables. A dataset of 90,000 training instances and 10,000 test instances was analyzed, with models including decision trees, neural networks, support vector machines (SVMs), k-nearest neighbor (KNN), naive Bayes, and logistic regression. Performance was evaluated based on accuracy, F1 score, AUC, training time, prediction speed, and model size. The results showed that the coarse decision tree achieved the highest accuracy (98.24%), the lowest error rate (1.76%) and the highest prediction speed (>420,000 observations/second) with the smallest model size (4.7 KB). The results underscore that simpler, easy-to-interpret models, such as decision trees, offer excellent practicality for real-time industrial applications without compromising predictive power. This work highlights the importance of balancing model complexity with computational efficiency in predictive maintenance systems.