Mortality modelling and forecasting: an application of the Lee-Carter model on the Japanese data

Accurate mortality modeling and forecasting are essential for understanding population health trends, informing public health policies, and improving demographic projections. This study applies the Lee-Carter (LC) model to Japanese mortality data from 1947 to 2022 to analyze trends and improve predictive accuracy. The dataset, obtained from the Human Mortality Database, is split into training and testing sets (90:10, 80:20, and 70:30) to evaluate model performance. To enhance projection accuracy, the LC model is integrated with the autoregressive integrated moving average (ARIMA) model. The findings reveal a steady decline in mortality rates for both genders, with female mortality following a smoother trend, while male mortality exhibits greater fluctuations. Life expectancy projections indicate a continued upward trend, with female life expectancy consistently surpassing that of males throughout the forecast period (2023–2032). Performance evaluations using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE) confirm that larger training datasets improve predictive accuracy, as demonstrated by the superior performance of the 90:10 split. The LC model effectively captures historical mortality trends, though some smoothing effects were observed. The U-shaped mortality pattern characterized by high infant mortality, low childhood mortality, and increasing mortality with age aligns with expected demographic trends. Additionally, economic recessions have had a stronger impact on male mortality, particularly due to increased suicide rates linked to job insecurity. Overall, this study demonstrates the LC model’s reliability in mortality forecasting, highlighting its potential for guiding public health policies. Future research may explore advanced machine learning techniques to further enhance predictive accuracy.