Research on fault prediction methods of medical electronic devices based on ARMA-RF models

• The nested ARMA-RF model reaches over 90 % accuracy in medical electronic device fault recognition (RF alone: >98 %, ARMA error: <5%). • ARMA prediction residuals retain structured info due to feature singularity in medical device electrical signals. • Voltage, current, and power signals of medical devices show > 0.999 similarity in dynamic patterns. • Feature-level redundancy limits the nested model’s discriminability despite its high overall accuracy. • The ARMA-RF framework provides a practical solution for real-time fault monitoring of medical electronic devices A combined ARMA-RF model based on Auto Regressive Moving Average (ARMA) prediction model and Random Forest (RF) classification model is adopted to realize the identification and prediction of the fault status of medical electronic equipment. Firstly, the electronic performance parameters current and voltage are selected as the original analyzed data, and a series of electronic signal feature extraction methods are proposed by observing the characteristics of the two signals in real measurement. Then, for the characteristics of smooth and periodic signals, the theoretical method of using RF as the identification model and ARMA as the prediction model is investigated, and the two models are fused into a system. After that, experimental validation can be designed to verify the recognition effect of RF and the prediction effect of ARMA model with the training set, and after the training is completed, the two models will be used in nesting and the prediction accuracy of the whole system will be verified using the test set. During individual training and testing, the recognition rate of RF model can reach more than 98%, when the average error of ARMA is below 5%; during nested training and testing, the final classification accuracy can reach more than 90%. Not only that, the RF model also ranked the weight of each feature. Medical electronic devices can be realized by a new machine learning model for their fault prediction, which provides a brand new idea for the realization of other electronic devices in fault prediction. Subsequently, this technique can be utilized to ensure the reliability of first aid equipment in many fields such as emergency rescue, wilderness rescue and medical support.