Modelling and Forecasting Mortality Rates Using a Neural Network-Integrated Lee-Carter Model

Artificialneural networks (ANNs), modelled after the brain's structure and function, can capture complex nonlinear relationships between predictors and response variables. This study integrates ANNs with the Lee-Carter (LC) framework using a multilayer feed-forward network to forecast the mortality index (kt), which tracks changes in mortality rates over time within the LC model. This mortality index is essential for forecasting future mortality patterns. Traditionally, the LC model uses an autoregressive integrated moving average (ARIMA) process to predict kt but ARIMA struggles with accurately forecasting future mortality trends. We compared the performance of the multilayer feed-forward network against ARIMA using mortality data from 19 countries, evaluating the models using root mean square error (RMSE) and mean absolute error (MAE). Our findings indicate that the multilayer feed-forward network outperforms ARIMA in forecasting mortality rates for 17 out of 19 countries. Additionally, integrating wavelet analysis and fuzzy logic with ANNs could further enhance forecasting accuracy by effectively managing non-stationary data, uncertainty, and complex patterns.