Mining metrological characteristics and predicting performance trends of syringe pumps based on K-Means and random forest

To address the limitations of current quality control for syringe pumps, which relies heavily on periodic verification and post-event assessment and lacks a prospective prediction mechanism for performance degradation trends, this study explores a machine learning-based method for metrological characteristic mining and performance prediction. Taking 284 single-channel syringe pumps from a tertiary-grade A hospital in Guangxi as research samples, the study collected key data on indicated error and repeatability under three typical flow rates, and constructed an analytical framework using K-Means clustering and Random Forest algorithms. The results indicate that indicated error displays significant differences between the two clustered groups, whereas repeatability shows no significant inter-group differences. Addressing the issue of high missed detection rates for abnormal samples caused by data imbalance, the Random Forest model was optimized using a cost-sensitive learning mechanism. This optimization significantly improved the recall rate for the performance-fluctuation group from 0.07 to 0.43, thereby enhancing the identification capability for potential high-risk devices. Although the overall prediction accuracy was adjusted to 62.8% due to increased safety redundancy, the model effectively reduces clinical missed detection risks. Future research should expand the sample size and enrich feature dimensions.