Health Risk Assessment And Prediction Based On Machine Learning

The aim of health risk assessment and prediction is to explore the potential health risk factors and to assess and predict the possible risk of the population and individuals.The implications are significant as it takes intervention measures for high-risk groups and improve the medical process.Around the health risk of population and individuals,four aspects of research have been conducted as follows:(1)We evaluated the effects of environmental factors on daily hospital admissions(HAs)for respiratory,circulatory,and digestive diseases.A generalized additive model(GAM)was utilized to control the confounding effects of seasonal trends and meteorological factors by using smoothing function,and to explore the association between air pollutants and hospitalization risk.We also performed subgroup and sensitivity analysis.Results showed that: respiratory and circulatory diseases are sensitive to air pollutants,the elderly(≥65 years)were more susceptible to acute exposure of pollutants.(2)We studied time series method of predicting the hospitalization volume related to environmental factors.We proposed a hybrid time series model(HTSM)based on nonparametric regression model and nonlinear residual fitting method.Nonparametric GAM was conducted to predict the HAs first,and then the residual is fitted by LSTM so that the prediction accuracy could be promoted.The model was tested on two datasets: daily HAs and weekly cumulative HAs.Results showed that HTSM can effectively predict the HAs for sensitive disease.(3)We studied unplanned readmission prediction method,and proposed an improved cost-sensitive integrated learning model.Samples of the minority class were enhanced to be learned by introducing cost sensitivity factors into AdaBoost algorithm.On the other hand,the overall classification accuracy was promoted by using dynamic cost search strategy.We also selected the important features and analysed the risk factors.Results showed that the improved model could effectively predict unplanned readmission data of actual medical scenario and the prediction accuracy have been improved as well.(4)We have constructed a health risk assessment and prediction system,integrated with original data,statistical model and machine learning algorithm,which can be used for early intervention of pollutant sensitive diseases,sensitive populations and high-risk readmission patients.This system provide an effective tool for the assessment and prediction of the health risks.