Machine Learning Techniques for Classifying Cardiac Arrhythmias

Cardiac arrhythmia, a condition characterized by irregular heartbeats, represents a significant health risk, contributing to 15–20% of deaths worldwide. These irregularities in the heart, whether tachycardia, bradycardia or any other condition, can be life-threatening if it is not diagnosed in time, and especially if it is not diagnosed correctly. Traditional diagnostic methods have the problem of the complexity of the nature of the electrocardiogram signals, as well as the variability of the data. These methods, in addition to being time-consuming, are prone to human error. In this study, we analyze three Deep Learning methods with the objective of improving the detection and classification of different types of arrhythmias. We used the Physionet MIT-BIH arrhythmia data set, which was divided into 80% training data and 20% test data. In preprocessing, we balanced the database and used a Butterworth low-pass filter to reduce noise and obtain only the part of the signal of interest. We compared three different architectures: a multilayer CNN, Mobile-Net and ResNet. The results obtained are very promising in terms of advances for rapid diagnosis of cardiac arrhythmias with accuracies ranging from 0.9779 to 0.9894, and very low losses between 0.0416 and 0.0652, depending on the model. The integration of these advanced models into real-time monitoring systems can provide immediate feedback and alerts for timely medical interventions, representing a powerful tool for preventive and personalized medicine.