ECG-based Daily Activity Recognition Using 1D Convolutional Neural Networks.

This study presents an approach to human activity recognition (HAR) using electrocardiogram (ECG) signals. We explore the application of ECG for not only providing cardiophysiological information but also for more extensive patient surveillance, including emergency recognition. Utilizing an end-to-end one-dimensional convolutional neural network (1D CNN), we analyzed wireless ECG data from 40 participants performing five common daily activities. Our research addresses the limitations of previous ECG-based HAR studies that depended on smaller, public datasets. By employing a 5-fold cross-validation (CV) and subject-independent methodology, our study ensures a broader applicability and generalizability of the HAR system. The model achieved a test accuracy of 82.9% and was particularly effective in recognizing the Sleeping activity with 98.5% accuracy. These findings not only validate the practicality of using ECG in HAR, but also open possibilities for comprehensive patient care, extending beyond traditional cardiac health monitoring.