Blood Pressure Value Prediction in Hemodialysis Patients Using the CNN+GRU Network Architecture

During clinical operation of Hemodialysis, Blood pressure (BP) variability, affect significant global risks and offshoot complication linked to adverse mortality. However, on this task, a prediction model has not been studied extensively with the advent of big data models. Earlier machine learning models lacked the exploitation of precise modeling regarding temporal context information affecting stratified BP values in the output. Recently deep learning models introduced hierarchical temporal feature engineering and context information for Spatio-temporal tasks, and thus this study geared to exploit the advantages of deep learning models, specially, in the form of the Convolutional Neural Network (CNN) model, variants of RNN models, and adaptive hybrid networks combining gated recurrent units (GRU) and CNN for the time-series tasks. In addition, a hybrid deep architecture of CNN and LSTM was employed to predict BP value and make it more efficient and effective when applied to health care. The benchmark results on our dataset for the validation of the CNN+GRU hybrid deep architecture model yielded promising contributions on the prediction task with most significant accuracy value of 72.7% corresponding to delta 15 mmHg and 83.7% when 20 mmHg.