ECG Data Analysis with MSE and Machine Learning

Time series data such as electrocardiogram (ECG) recordings display patterns that indicate regularity in heartbeat rhythms. Medical researchers normally analyze data statistically after collecting all data, but this kind of post-processing approach is difficult to be applied to real-time issue, such as identifying abnormal heartbeats. To address this issue, we investigated applying supervised machine learning approaches to RR intervals of ECG data for classifying three different types of heartbeat patterns: Healthy Heartbeat, Heart Failure and Atrial Fibrillation. In our study, we also applied the same methods to Multiscale Entropy (MSE) of RR intervals to verify if we can have a better prediction result. The evaluation results show that simply using RR Intervals of ECG data can build machine learning models to predict unseen ECG data with an 80%+ accuracy. However, there is no significant improvement when using MSE of RR intervals. The findings of this study demonstrated that supervised machine learning approaches are relatively successful in predicting heartbeat types with RR interval data than using MSE technique of coarse graining data with higher time scales. Based on this finding, we believe there is a chance to build a mobile or Web-based system that can use a trained model to distinguish three heartbeat patterns for real-time service.