Short-term heart rate prediction approach based on CNN-GRU model with an attention mechanism

Abstract

An accurate short-term heart rate (HR) prediction approach can provide safe, reliable, and efficient early warning for human health, and reduce the occurrence of harmful events. Traditional heart rate prediction methods cannot meet the requirements of high-precision dynamic prediction, and the wide application of machine learning algorithms provides a series of accurate methods for HR short-term prediction. In this study, one kind of a hybrid model, the CNN-GRU model with an attention mechanism, for HR prediction is designed, which is facilitated by the convolutional neural network (CNN) as well as the gated recurrent unit (GRU) with an attention mechanism. This model can realize more precise prediction and early warning for short-term HR changes with full consideration of its dynamic features. To prove the accuracy of our forecasting method, the college students’ HR trends in their daily life are acquired by the wireless HR monitoring devices. This hybrid prediction algorithm is been proved to be effective when compares with other commonly-used prediction models (CNN, LSTM, GRU, SVM, Random Forest). Through experiments, our proposed model is proved to has higher accuracy in HR prediction and early-warning application.