Enhancing Machine Failure Detection with Ensemble Learning and Oversampling Techniques

In modern manufacturing, predictive maintenance plays a vital role in minimizing unexpected machine failures, thereby reducing downtime and maintenance expenses. This study explores the implementation of ensemble learning models, including Bagging, AdaBoost, Gradient Boosting, Random Forest, and XGBoost, to predict machine failures based on operational characteristics such as air temperature, process temperature, rotational speed, torque, and tool wear. To address the class imbalance commonly present in failure datasets, oversampling techniques like SMOTE (Synthetic Minority Over-sampling Technique) and ADASYN (Adaptive Synthetic Sampling) were used to improve the model’s performance. However, a couple of experiments shows that ensemble models without oversampling is better than those that uses these techniques. The Gradient Boosting model using just the oversampling performed best with an F1-score of 0.78 for class 1 and overall accuracy of 0.99 while specifically, the version without any oversampling made it to an F1 Score of even as high as 0. XGBoost and Bagging were the next best models with F1-scores of 0.76 and 0.75, respectively since then. In contrast, the F1-score of the model proposed in this study and the models using any kind of oversampling technique were lower than 0.70 of failure class. This is proof of the power of ensemble methods (especially Gradient Boosting), that as result can be deployed to forecast machine failures and encourage right preventive maintenance actions ensuring less interruptions and smoother operations.