Classification of Physical Activity from Electrocardiogram Signals Using Deep Learning

Accurate classification of physical activity based on ECG signals is essential for monitoring health and improving intervention strategies. This study focuses on utilizing ECG signals for the classification of physical activities through deep learning techniques. The analysis involves processing ECG time series data into Continuous Wavelet Transform (CWT) based scalogram images. ECG recordings are sourced from a dataset that represents various activity states: during mathematical problem-solving (M), while sitting (S), during hand-biking (H), and during walking (W). Convolutional neural network (CNN) architectures such as MobileNetV2, EfficientNetB0, DenseNet201, VGG16, and ResNet152 use CWT-based scalogram images as their input. MobileNetV2 shows the most effective performance in classifying these activity states, achieving testing, and validation accuracies of 97.07%, and 97.83%, respectively. The average of the performance parameters: Precision, Sensitivity, Specificity, and F1-scores are 97.75%, 97.75%, 99.25%, and 97.75% respectively. The results demonstrate the potential of deep learning in enhancing the classification of physical activities based on ECG data.