Tree-Based Learning Algorithms to Classify ECG with Arrhythmias

An arrhythmia is an abnormality in heart rate or rhythm. Depending on which part of the heart is affected and its abnormality pattern (too fast, too slow or irregular), arrhythmia can be classified into around a dozen of classes. Accordingly, arrhythmia treatments vary depending on the type of arrhythmia. An electrocardiogram (ECG) records the electrical signals in the heart to profile the heart rate, rhythm, and patterns, which collectively allows the detection and diagnosis of different types of arrhythmias. Inconsistent and insufficient accuracy of ECG interpretation by medical professionals is a problem in clinics for which statistical algorithms can be a promising alternative. In this study, we explored four learning algorithms, gradient boosting, random forest, decision tree and extra trees, to classify arrhythmia on ECG signals from over 10,000 patients. Specifically, we extracted features from ECG signals, evaluated four tree-based learning on classification of arrhythmia and evaluated each learning algorithm’s performance in terms of accuracy, precision, recall, and F1 based on 10-fold cross-validation. We hypothesized that gradient boosting would be able to classify arrhythmia with high accuracy when trained. Here we report that gradient boosting provided an accuracy of 95%, outperforming all other classification models in all the performance measures evaluated in this study. The findings here suggest that the statistical classification using the data science tools could greatly facilitate ECG detection and diagnostics.