Melika Mirzaseyedi

and 2 more

Today, heart disease is one of the most common causes of death in the world. For this reason, accurate and timely diagnosis of patients' conditions is essential. Because traditional diagnostic methods for these kinds of diseases have many costly side effects, researchers are always looking for cheaper and more accurate ways to diagnose. One effective strategy is using mobile health devices to monitor the person's health status through the signals received by an electrocardiogram. Early diagnosis enables more effective treatment and prevention of chronic diseases. Measuring and recording electrocardiogram signals by mobile services is a valuable solution in the field of health care, such as processes for classifying and diagnosing the activity or condition of patients. This study aimed to find a model with a precise function for classifying and recognizing human activities using cardiac data taken from wearable sensor signals from two electrocardiography sensor. This classification model of human activity is based on the electrocardiogram data of the m-health data set and obtained using four deep learning models. These models include recurrent neural networks like (LSTM), Convolutional Neural Network (CNN), and attention-based hybrid neural networks. This research has taken a step in classifying different types of human activities only through electrocardiogram signals. By examining other hyper parameters and selecting the best ones, our proposed network has achieved slightly higher accuracy than previous research. The proposed CNN-LSTM combination-based learning model obtained approximately values of 100% accuracy, 100% precision, 100% recall, 100% F1 score, and 100% for the ROC AUC scale, respectively, in terms of validation results. The results show that our designed network performs better than the advanced models of previous research.