Development of a patient health monitoring system based on the internet of things with a module for predicting vital signs

Abstract

Recent issues related to human health in the world have shown the importance of telemedicine considering necessities to perform the remote monitoring of patients. In this study, using a patient smart monitoring system (PSMS), we collected 5,000 samples of heart rate and blood saturation vital signs from 4 volunteers and tried to find better correlation algorithms to develop a module to predict what these vital signs will be in the next 60 seconds. The following regression algorithms recurrent neural network (long short-term memory) (RNN(LSTM)), autorregresive integrated moving average (ARIMA), value-added reseller vector autoregression (VAR) were used to forecast the patient's state of health in the next 60 seconds. Further, the support vector machine (SVM) and Naive Bayes classification algorithms use the data forecasted by the regression algorithms as input to predict the health status of the patients. When comparing algorithms, we focused on the F measure, a metric used to evaluate the performance of machine learning algorithms. As a result, RNN(LSTM) and SVM showed the performance score value of machine learning algorithms F 0.84, RNN(LSTM) and Naive Bayes 0.81, VAR and SVM 0.82, and VAR and Naive Bayes 0.75. Compared to them, the correlation of ARIMA regression algorithms and SVM classification showed a better F score of 0.86 for machine learning algorithms than the others.