This is a project made for a Human Data Analytics class during a Master's degree course in ICT.
Abstract
The automatic classification of heart rhythms using short time single lead ECG recordings is a challenging task that has been widely studied recently. In this paper we present our work that aims at classifying these kind of ECG signals as Atrial Fibrillation (Afib), Normal, Other rhythms or too noisy to be classified (Noisy). We developed three different binary classifiers as Recurrent Neural Networks (RNNs) both with a binary cross-entropy loss function and a weighted version of it. We used these three RNNs to develop a cascade classifier for the samples of the given dataset, considering the problem as a multiple binary classification problem. We obtained similar results, with a slightly better result using the unweighted loss function, with an accuracy of 81.18% vs 80.01% and a F1 score of 0.77 vs 0.76.
Hierarchy of the process blocks.
Final classifiers hierarchy.
Validation performances of the single trainedmodels.
The final accuracy of our project is 81.18% with a F1 score of 0.77
For the paper, the license used is: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
