Arrhythmia detection using amplitude difference features based on random forest

A number of promising studies have been proposed for diagnosing arrhythmia, using classification techniques based on a variety of heartbeat features by the interpretation of electrocardiogram (ECG). In this study, a new feature called amplitude difference was investigated using the random forest classifier. Evaluations conducted against the MIT-BIH arrhythmia database before and after adding the amplitude difference features obtained heartbeat classification accuracies of 98.51% and 98.68%, respectively. To validate the significance of the increased performance, the Wilcoxon signed rank test was extensively employed. By the absolute preponderance of plus ranks, we confirmed that applying an amplitude difference feature for heartbeat classification improves their performance.