Mortality Forecasting Of Chinese Elderly Population Based On Neural Network

Modeling and forecasting mortality in the aging population is fundamental to measuring longevity risk and managing pension risk.On the one hand,in my co untry,the death data of the elderly population of retirement age and above are scarce,and the random fluctuations are large.It has always been difficult to accurately predict the mortality rate of the elderly population in the long run.If the mortality model is established by gender,the predicted mortality values of women and men are prone to produce results that are contrary to the laws of human social development.On the other hand,with the continuous development of data science,machine learning methods such as neural networks have been gradually introduced into the field of mortality modeling,and it has been confirmed that they can improve the prediction effect of mortality.In previous studies on mortality prediction of elderly population in my country,mostly based on traditional mortality models,and few studies have applied machine learning methods to mortality prediction in the elderly population.In order to further improve the prediction effect of the mortality rate of the elderly population in my country,from the perspective of deep learning,this paper considers the time series characteristics of the mortality rate data of the elderly population,and uses the Embedding layer to combine gender variables to build Convolutional Neural Network(CNN),Simple Recurrent Neural Network(Simple RNN),and Long Short-Term Memory(LSTM)Mortality Models for fitting and forecasting gender-separated mortality of the elderly population jointly.Divide the mortality data of the elderly population of men and women in my coun try,and use the data from 1995 to 2015 as the training set.Apply each neural network mortality model to fit the data of the training set,and obtain the predicted value of the mortality rate of the elderly population of men and women from 2016 to 2050.Comparing with Logistic two-population mortality model in terms of fitting and prediction,the application of neural network model to mortality data of elderly population in my country is discussed.The final result shows that although the logistic two-population model has better fitting effect,for a prediction model,we pay more attention to its performance on unknown data.CNN model and Simple RNN model have higher prediction accuracy than the traditional logistic two-population model.After the biological rationality test,it is found that the Simple RNN model is more suitable for the mortality data of the elderly population in my country,and its male-female mortality ratio will converge to a fixed value over time,satisfying the reality that the mortality rates will converge in the interrelated population.Finally,the mortality data of the population by sex and age predicted by the Simple RNN model are used for calculation,and it is concluded that the life expectancy of men and women in my country’s 60-year-old population will gradually increase over time,and the life expectancy of women is higher than that of men.