Research On Population Mortality Based On Machine Learning Methods

With the improvement of living standards and the improvement of medical and health conditions,the mortality rate of our country’s population has been declining year by year,life expectancy has gradually increased,and the degree of population aging has continued to deepen.The longevity risk brought by population aging has become a social problem that cannot be ignored.Longevity risks increase the future solvency pressure of the pension insurance system and are one of the important risks affecting the sustainable development of the pension insurance system.Constructing an appropriate mortality model that can accurately predict the future mortality rate of the population is the basis for quantifying longevity risks,and it plays an important role in the government’s formulation of relevant pension policies.Therefore,research on population mortality is of great significance in the fields of longevity risk management and population policy formulation.In order to improve the prediction effect of my country’s population mortality rate,the article is based on the mortality data of China and the United States over the years,using the classic Lee-Carter model as the benchmark model,and then based on the machine learning method to study the population mortality model,and compare the differences.Machine learning methods can improve the effect of data fitting,thereby improving the effect of predicting the mortality rate of our country’s population.The specific work of the article is as follows: First,briefly describe the basic situation of crude death rates in China and the United States.Then,based on the mortality data of all age groups in China and the United States,the data set is divided into a training set and a test set,and a representative mortality model Lee-Carter model as an example,starting from the assumption that the number of deaths obeys the Poisson distribution,a Poisson regression tree,random forest and neural network models are constructed to minimize the "deviation statistics" to improve the model fitting effect,analyze the improvement effect,and The detection model is insufficient in a certain aspect,and the direction for the improvement of the model form is selected.The results show that decision trees,random forests and neural network models can significantly improve the model fitting effect.For the mortality data of China and the United States,both the internal and external losses of the sample show that the neural network model has a better fitting effect than the regression tree and the random forest model,which means that the neural network model is more suitable for the modeling and analysis of my country’s population mortality.