Data Extraction Method Combined with Machine Learning Techniques for the Detection of Premature Ventricular Contractions in Real-Time

Abstract

Currently, diagnostics in the medical field are being automated. Thus, reducing errors of interpretation in diagnoses. This article proposes a recognition method to identify premature ventricular contraction in real time, soon enabling the minimization of damages resulting from arrhythmia. The proposed method consists of two main modules: data extraction module by way of Recursive Least Squares (RLS), guaranteeing data extraction in real time, and the classification module, its inputs being the parameters from the RLS algorithm. In the resource extraction module, autoregressive modeling (AR) is used to extract characteristics. In the classifier module, Support Vector Machine and Multilayer Perceptron are examined. The classifiers’ performance was assessed by standard metrics. The proposed algorithm showed high precision and few false negatives.