Impact of machine learning on spare parts availability for critical medical devices: a supervised machine learning...

This paper seeks to improve the reliability and quality of operation of the critical medical equipment methods through the combination of failure mode and effects analysis (FMEA) and supervised machine learning (ML) approaches to predictive spare parts management. The paper aims to reduce downtime, optimise the maintenance planning and increase the quality of healthcare services in general by advanced decision support.

A dataset comprising 2,800 maintenance records from six hospitals, covering 10 categories of medical devices including ventilators, dialysis machines, infusion pumps and computed tomography scanners, was analysed. The FMEA was initially used to calculate risk priority numbers (RPNs) to indicate the criticality of devices and their probability of failure. These RPNs were used as input features to three supervised ML models, which are the random forest (RF), the artificial neural network (ANN) and the support vector machine (SVM). Each model underwent grid-search hyperparameter tuning and five-fold stratified cross-validation to ensure reproducibility. The accuracy, precision, recall, F1-score and area under the curve were used to evaluate the models.

The accuracy of the RF model and ANN was 1.00 with an F1-score of 0.90 and SVM had the highest recall of 0.94, implying that it is more sensitive in identifying actual spare parts replacement requirements. The combined failure mode and effects analysis-machine learning model enhanced the availability of spare parts by an estimated 12-15% in all hospital locations. The strength and discriminative performance of the proposed models were tested with the help of visual analysis through confusion matrices and receiver operating characteristic curves. These outcomes demonstrate how predictive analytics can strengthen maintenance traceability and ensure continuity of patient-care equipment.

The suggested hybrid structure assists in data-based maintenance planning within the biomedical engineering departments of hospitals. It helps minimise the downtime, promote patient safety and increase the adherence to the healthcare quality standards, including International Organization for Standardization 13485 and Joint Commission International accreditation, by predicting replacement requirements and aligning them with inventory management. The model is also consistent with United Nations Sustainable Development Goals (SDGs) (SDG 3: Good Health and Well-Being and SDG 12: Responsible Consumption and Production).

Hospitals can optimise inventory planning and preventive maintenance scheduling.

Improved equipment availability supports better patient care and healthcare reliability.

This study is one of the pioneering studies that combine FMEA-based criticality measures and supervised ML models (RF, ANN and SVM) to optimise hospital maintenance. It offers a clear and understandable system that connects traditional reliability engineering and intelligent decision-support systems, which gives quantifiable gains in both healthcare quality assurance and sustainability performance.