Analysis And Prediction Of Time Series Health Data

With the rapid development of medical informatization and wearable devices,more and more health data can be obtained from electronic medical records or wearable devices.In the medical field,many medical and health data exist in the form of time series,such as continuous monitoring of blood glucose,blood pressure,and blood lipids associated with chronic diseases,continuous monitoring of heartbeat,pulse,and posture changes by smart wearable devices.There are two main problems in the process of actually constructing a time series health data prediction model.One is to perform prediction on continuous time series data.The result of the prediction is the data of the time series data at a future point in time,which is called continuous time series data prediction problem.Another type of problem is the use of time series data and related cross-sectional data to predict the relevant data at a specific point in the future,which is called the prediction problem of time series feature guidance.This thesis focuses on the above two issues for further study,and the main work is as follows:(1)For the prediction problem guided by time series features,this thesis proposes a prediction model(TS-DNN)guided by time series features and based on deep neural networks.The model considers time series data and related cross-section data as two types of neural network input.It corrects the model's prediction results and improves the model's prediction accuracy through a deep learning of the potential trend of time series data.(2)For the continuous time series data prediction problem,this thesis constructs a hybrid model of ARIMA-SVM and combines the advantages of ARIMA and SVM models to improve the accuracy of continuous time series data prediction.The hybrid model first uses the classical time series forecasting model ARIMA to model the linear part of the time series data.The prediction residuals of the ARIMA model are predicted by the SVM,and the results of the two parts are added together to obtain the final prediction result,thereby completing the continuous prediction of time series data.(3)In this thesis,the above two models are applied to the scene of newborn weight prediction and continuous blood pressure prediction respectively.In the prediction of newborn weight,the serial data of the weight of pregnant women and the physiological parameters of the pregnant women and the fetus are taken as the input of the model,and the compared with the traditional formula prediction method and the in-depth neural network prediction model using single-time cross-section data,TS-DNN is more accurate and improves the stability of model prediction results.In the problem of continuous blood pressure prediction,the respective advantages of the ARIMA and SVM models are fully utilized by using a hybrid model.The experimental results show that the hybrid model can more accurately capture potential changes in continuous blood pressure time series data and improve blood pressure prediction and provide technical support for future risk warnings.