Adult and Non-Adult Classification Using ECG

Abstract

In this paper, we proposed an age classification scheme using Electrocardiogram (ECG). We experimented using some datasets from the PTB-XL ECG database, which were created in accordance with the age distribution of the Canadian population. We specifically focused on the QRS portion of the ECG wave as an indicator of age. Our scheme contained band-pass filtering; discrete wavelet decomposition reconstruction; a deep neural network having 1D CNN, LSTM, and regression layers. Initially, age prediction was attempted using this method. However, predicted ages were found to center around a certain region at the start of adulthood. This result prompted us to explore age classification. For age classification (adults and non-adults), our method achieved classification accuracies up to 99%. Such promising outcomes generated the feasibilities of further experimentation and possible practical implementation of ECG for anonymous age verification.