Electrocardiogram Feature Based Heart Arrhythmia Detection Using Machine Learning and Apache Spark

Abstract Heart arrhythmias are the main cause of death worldwide. Electrocardiogram (ECG) results can be used to identify arrhythmias, or irregularities in the heart's rhythm. Because symptoms are not always present, the diagnosis is often off. To prevent a potentially catastrophic situation, patients using real-time ECG monitoring must identify arrhythmias early on. In this work, Structured Streaming, an open-source Apache Spark technology, was used. Finding a method to apply machine learning to detect cardiac arrhythmias in real-time is the goal of the project. Investigating how structured streaming affects metrics for content classification and how long it takes to find episodes was another goal. At MIT and BIH, we have been gathering ECG information. With this information, arrhythmias like RBBB and atrial fibrillation might be recognised and categorised. There are many methods for separating these erratic rhythms from one another. We used a multiclass classifier based on logistic regression, a random forest, and three different decision trees to categorise the data. The random forest classifier wins out when the three classification methods are compared. In comparison to other studies, this study demonstrated improved classification model performance metrics and decreased pipeline runtime.