Densely Connected Neural Network and Permutation Entropy in the Early Diagnostic in COVID Patients

The COVID-19 pandemic has been characterized by the high number of infected cases due to its rapid spread around the world, with more than 6 million of deaths. Given that we are all at risk of acquiring this disease and that vaccines do not completely stop its spread, it is necessary to continue proposing tools that help mitigate it. This is the reason why it is ideal to develop a method for early detection of the disease, for which this work uses the Stanford University database to classify patients with SARS-CoV-2, also commonly called as COVID-19, and healthy ones. In order to do that we used a densely connected neural network on a total of 77 statistical features, including permutation entropy, that were contrasted from two different time windows, extracted from the heart rate of 24 COVID patients and 24 healthy people. The results of the classification process reached an accuracy of 86.67% and 100% of precision with the additional parameters of recall and F1-score being 80% and 88.89% respectively. Finally, from the ROC curve for this classification model it could be calculated an AUC of 0.982.