Fault diagnosis in rolling element bearings is critical for ensuring machinery reliability. This study improves machine learning techniques for predictive fault detection using the benchmark CWRU bearing dataset. Vibration signal data is preprocessed via balancing and graph-based feature engineering is performed to enable effective model training. Diverse classifiers including Random Forests, Support Vector Machines and Neural Networks are systematically evaluated through 10-fold cross-validation. Most of the models demonstrate exceptional performance, with top accuracies and AUC scores of 1.00. The research highlights the potential of hidden features that consider the implicit relations between the entities to improve predictive maintenance through data-driven bearing fault diagnosis.
