Research on Intelligent Calibration Test Fault Diagnosis Model of Automatic Chemiluminescence Immunoassay Analyzer

In this study, three fault diagnosis models for the calibration test of the automated chemiluminescent immunoassay analyzer (ACLIA) are developed. The models are trained, validated and tested with a large amount of historical calibration testing data. We compare the performance of three machine learning methods of Random Forest (RFC),Extreme Gradient Boosting (XGB) and Light Gradient Boosting Machine (LGBM). The results on independent test datasets show that the RFC model performs the best, with accuracy, F1, and AUC metrics of 0.979, 0.982, and 0.930, respectively; the XGB model performs similarly to the RFC model, with accuracy, F1, and AUC metrics of 0.964, 0.974, and 0.910, respectively; the LGBM model has the AUC metric of 0.900. In summary, the RFC,XGB and LGBM models trained in this study can effectively diagnose the calibration fault categories of ACLIA, which can help medical instrument manufacturers reduce the losses caused by the calibration testing process, reduce the cost of reagents, and improve the production efficiency.