A Comparative Study of Machine Learning Algorithms for Predictive Maintenance in Manufacturing Systems

Abstract The global manufacturing sector is rapidly transitioning towards smarter, data-driven operations to enhance productivity, reduce operational costs, and improve overall equipment effectiveness. Predictive Maintenance (PdM) stands out as one of the most transformative approaches within this paradigm. By anticipating equipment failures before they occur, PdM ensures that maintenance is carried out only when required, leading to optimal utilization of resources. Machine learning (ML) techniques form the core of PdM, leveraging data to detect patterns and anomalies that precede system failures. This paper presents a comprehensive comparative study of leading ML algorithms—both traditional and deep learning-based—deployed for PdM in manufacturing systems. Algorithms such as Decision Trees, Random Forests, Support Vector Machines, K-Nearest Neighbors, Multilayer Perceptrons, Long Short-Term Memory (LSTM) networks, and Convolutional Neural Networks (CNNs) are rigorously evaluated on standardized datasets. Each model is analyzed in terms of prediction accuracy, interpretability, computational efficiency, and suitability for deployment in industrial settings. The results reveal key trade-offs and offer recommendations tailored to different types of manufacturing environments. The paper also discusses the current challenges in PdM adoption and offers insights into future research directions, including the integration of explainable AI and real-time edge computing solutions.