Predictive maintenance is widely used in modern industrial systems. It helps improve the working life of machines. It also reduces risk and lowers overall operating costs. Many current approaches still face problems when handling both fast processing and balanced performance across different measurements. In this research, twelve machine learning models are tested. These include standard algorithms and deep learning-based solutions. Two manufacturing datasets are used. One has more samples, while the other shows uneven class labels. Important features are selected by applying strict screening. Model parameters are fine-tuned to obtain stable results. To measure how each model performs, several metrics are used—accuracy, precision, recall, F1-score, and ROC AUC. Among all tested models, random forest shows the best results. It reaches a classification accuracy of 99.5%. At the same time, it keeps a good balance between recall and precision. This model works well when data from sensors is imbalanced. It is also strong in handling patterns that do not follow a clear rule. The system is potentially suitable for real-time deployment in industrial machines with rotating parts, as demonstrated on two representative manufacturing datasets. However, broader validation across diverse equipment types is recommended before large-scale adoption.
Yang et al. (Mon,) studied this question.