Application Of Machine Learning In Specific-cause Mortality Prediction

With the rapid development of economy and the continuous improvement of medical level,the life expectancy of the population continues to extend.Behind the rapid increase of life expectancy is the decline of overall mortality,which is determined by the mortality of different causes of death.Therefore,this paper takes the death cause mortality rate as the research object,referring to the International Classification of Diseases(ICD-10)and existing studies,and combining with the current death cause structure in China,divides the death cause into five categories : tumor,cardiovascular and cerebrovascular diseases,respiratory and digestive diseases,injuries,and others.The mortality rates of various causes of death were predicted and analyzed,and the impact of future causes of death on life expectancy was explored,so as to provide the basis for the relevant departments to reasonably allocate public health resources and formulate correct disease prevention and control plans.At the same time,machine learning has developed rapidly and become a research hotspot in many fields.In the past mortality prediction,the time term of mortality model is mostly extrapolated by traditional ARIMA method.In view of the limitations of ARIMA,this paper introduces two machine learning methods,artificial neural network(ANN)and support vector regression(SVR),to model and predict the time factor kt of Lee-Carter(LC)model,and constructs the ’ machine learning based mortality prediction model ’ : LC-ANN and LC-SVR,and compares the prediction results with ARIMA.Specifically,taking the LC model as the starting point,the parameters of the model are estimated,and the age factors ax,bx and time factor kt are obtained.The kt value is divided into the training set within the sample(1990-2014)and the test set outside the sample(2015-2019).After analysis,it is found that the fitting effect within the sample is ANN > SVR > ARIMA,and the prediction effect outside the sample is SVR > ARIMA > ANN.Although the fitting accuracy of SVR within the sample is slightly inferior to that of ANN,the prediction accuracy outside the sample is higher than that of ANNand ARIMA,which has good generalization ability.The generalization is the primary factor to evaluate the quality of a prediction model.Therefore,we use SVR to predict the kt value of each death cause mortality from 2020 to 2029,and replace the predicted kt value to the LC model expression to obtain the mortality of each death cause age in the next decade.Finally,based on the predicted mortality rate,the death cause life table was prepared for death cause analysis.The analysis results show that after removing cardiovascular and cerebrovascular diseases,tumors,respiratory and digestive diseases and injuries in 2019,the increase in life expectancy is 7.46,3.09,1.44 and 1.13 years old,respectively,and that in 2029 is7.94 and 3.31,1.06 and 0.92 years old,respectively.To a certain extent,extent that the death threat of tumors and cardiovascular and cerebrovascular diseases to the population is deeper than that in the past,and the death threat of respiratory and digestive diseases and injuries to people is weakened.Therefore,the health sector can consider increasing the supply and allocation of health resources for tumors and cardiovascular and cerebrovascular diseases.