A comprehensive assessment of machine learning models for predictive maintenance using a decision-making framework in the industrial sector

In the context of Industry 4.0, predictive maintenance models play a key role in improving operational efficiency and minimizing downtime in manufacturing systems. This study offers a comprehensive evaluation of five machine learning models. Random Forest, Gradient Boosting, K-nearest neighbors, Support Vector Machines, and Decision Trees by examining their performance based on critical criteria. These criteria include interpretability, accuracy, robustness, scalability, training time, and computational efficiency. The CRITIC method is used to calculate the objective weights for each criterion, leveraging their relative importance. Subsequently, the alternative ranking order technique based on spherical fuzzy sets (SFSs), known as the AROMAN approach, is applied to rank the models. To address the uncertainties inherent in real-world applications, various aggregation operators (AOs) are developed and utilized. The findings reveal that Gradient Boosting outperforms the other models, demonstrating excellent predictive performance and computational efficiency, making it a strong candidate for predictive maintenance applications. The study underscores the importance of model interpretability and trustworthiness in machine learning. Additionally, it suggests future research directions, such as improving Gradient Boosting techniques or creating hybrid models that combine the strengths of multiple algorithms. In general, this research serves as a valuable resource for practitioners and researchers aiming to advance predictive maintenance strategies in manufacturing environments.