Desenvolvimento de um modelo de aprendizado de máquina para detecção de arritmias cardíacas em eletrocardiogramas

Abstract

Cardiovascular diseases are a leading cause of morbidity and mortality worldwide, with cardiac arrhythmias being one of the most common and challenging disorders to diagnose. Automated analysis of cardiac signals has become an essential tool to aid in the diagnosis of these conditions, providing greater efficiency and accuracy in the detection of anomalies. This work aimed to develop a machine learning model for the classification of heartbeats from electrocardiogram signals, using AutoML techniques to optimize the choice of models and hyperparameters. The MIT-BIH database used contains multiple classes of heartbeats, of which five main ones were selected for this study. The methodology involved stages of database selection, data preprocessing, model generation and model evaluation. The Random Forest BAG L2 algorithm presented the best performance, with 98.39% accuracy and 98.34% F1-score, standing out as a robust and efficient model for classification. During development, challenges related to class imbalance and similarity between certain types of heartbeats were faced, which were mitigated through hyperparameter tuning and data normalization techniques. The study showed that the methodology employed was effective in building a reliable model for analyzing cardiac signals.