Predicting cardiovascular events with deep learning approach in the context of the internet of things

Nowadays, one of the causes leading to death in all over the world is the occurrence of arterial and cardiovascular events, which results in heart failure and premature deaths occurring in the form of myocardial infarction, stroke, and fainting. Therefore, it is essential to inform people before disasters occurring to prevent and warn of abnormal conditions. In this paper, a deep learning approach was used to predict arterial events over the course of a few weeks/months prior to the event using a 5-min electrocardiogram (ECG) recording and extracting time-frequency features of ECG signals. Considering the possibility of learning long-term dependencies to identify and prevent these events as quickly as possible, the Long Short-Term Memory (LSTM) neural network was used. A Deep Belief Network (DBN) was also used to represent and select more efficient and effective features of the recorded dataset. This approach is briefly called LSTM-DBN. Four publicly available datasets in the field of health care were used to evaluate the proposed approach. These data were collected from wearable heart rate monitoring sensors along with demographic features in the context of the Internet of Things. The prediction results of the proposed LSTM-DBN were compared with other deep learning approaches (simple RNN, GRU, CNN and Ensemble), and traditional classification approaches (MLP, SVM, Logistic Regression and Random Forest). In addition, DBN performance was compared with other methods of feature selection and representation such as PCA and AutoEncoder. Experimental results showed that the proposed LSTM-DBN (88.42% mean accuracy) had significantly better performance in comparison with all other deep learning approaches and traditional classifications.