A Survey on Cardiac Arrest Detection Using Machine Learning
In hospitals, most IHCA cases are preventable, but the survival rate of In-hospital cardiac arrest (IHCA) patients decreased. More than 54% of IHCA patients had irregular clinical symptoms that had previously gone into cardiac arrest. Take appropriate measures before the IHCA increases patient survival rates and reduces medical expenses. In this paper, a new method for diagnosing events prior to IHCA occurs. Build a dual-shift window (two tasks) that can apply machine learning to a very disproportionate dataset. The results indicate that this method can successfully process an unbalanced dataset to detect cardiac arrest. With the area under the selection performance line, the Area Under the Receiver's Operating Characteristic Curve (AUROC) and Accuracy Under Precision Recall Curve (AUPRC), the finest classifiers are random forests used for task 1 and AUROC of 0.88. LSTM is best for task 2, and the AUPRC for the next task is 0.71. Use the resampling technique to adjust the amount of data between CPR and non-CPR patients in the dataset, clean the data, and build a sliding window. Reduce datasets by applying multiple classifiers to model training. Excessive problems may occur. In addition, the performance of the model is compared and measured using the operating characteristics of the receiver, such as the operating characteristics of the receiver (e.g., the area under the operating characteristics curve (AUROC) of the receiver, and the area under the accuracy evaluation curve (AUPRC).