A SYSTEMATIC REVIEW OF FAULT DETECTION IN MEDICAL EQUIPMENT USING MACHINE LEARNING: ALGORITHMS, PERFORMANCE, AND SCALABILITY CHALLENGES

Unexpected failures of medical equipment disrupt healthcare services, compromise patient safety, and increase operational costs. Predictive maintenance powered by machine learning (ML) offers a promising solution, but the effectiveness of different ML algorithms across various medical devices and data scales remains unclear. This systematic review aims to identify, analyze, and synthesize the existing literature on ML-based fault detection and failure prediction in medical equipment, with a special focus on the types of algorithms used, their reported performance, and the characteristics of datasets employed. A structured search was conducted across five digital libraries (Web of Science, ACM Digital Library, ScienceDirect, Wiley Online Library, IEEE Xplore) for studies published between January 2018 and March 2025. Studies were included if they applied ML algorithms to detect or predict faults in medical equipment and reported empirical results from real or realistically simulated medical data. Data extraction focused on problem type, dataset size, algorithm(s), accuracy/metrics, and tools. Eleven studies met the inclusion criteria. Support Vector Machine (SVM) and its variants were the most frequently used algorithms for nonlinear fault patterns. SVM achieved high accuracy (often >90%) on small to medium datasets (≤13,000 records) but showed performance degradation on larger datasets. Ensemble methods, particularly with AdaBoost, achieved the highest accuracy (79.5%) on a large multi-institutional dataset (8,294 devices). ML, especially SVM and ensemble methods, can effectively detect faults in medical equipment, but model generalizability and scalability to big data remain significant challenges. Future research should focus on hybrid approaches that combine nonlinear SVMs with dimensionality reduction techniques (e.g., Gaussian pyramid) to maintain accuracy while handling large-scale, real-world IoT data from medical devices.