Forecasting youth mortality in Zimbabwe using HYBRID LEE-CARTER MODELS: A comparative analysis of Arima and Random Forest Approaches.
- Author
- Verengai, Ashleigh T.
- Title
- Forecasting youth mortality in Zimbabwe using HYBRID LEE-CARTER MODELS: A comparative analysis of Arima and Random Forest Approaches.
- Abstract
- Youth mortality remains a critical public health issue in Zimbabwe, particularly for youth aged 0-24 years. Accurate mortality forecasting is vital for informing evidence-based health interventions and policies. In this research, there is a comparative evaluation between two hybrid models for mortality forecasting: the hybrid Lee-Carter model with Auto-Regressive Integrated Moving Average (LC-ARIMA) and the hybrid Lee-Carter model with Random Forest regression (LC-RF). Mortality rate data from 1990-2011 was obtained from UNICEF and Singular Value Decomposition was used to estimate parameters for the model. Estimates for the period 2012-2022 were generated by utilizing ARIMA and Random Forest techniques to estimate the time-varying mortality index (๐๐ก). Quantitative evaluation of the models was performed using error metrics Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Percentage Error (RMSPE). Results show that the LC-RF model performed better in projecting mortality for age brackets 0โ4, 5โ9, and 20โ24 years, where it effectively captured non-linear trends. Conversely, LC-ARIMA fared better in the 10โ14 and 15โ19 age brackets. The results demonstrate the value of hybrid model approaches and affirm the necessity to apply suitable models to the underlying trends in the data. This study contributes to the body of literature by applying hybrid mortality models to a relatively understudied age cohort in Zimbabwe and offers results relevant for public health planning, actuarial purposes, and future mortality research. For areas of further studies, the researcher could use other predictors such as socioeconomic or health system predictors. Other machine learning algorithms such as LSTM and XGBoost can be used.
- Date
- June 2025
- Publisher
- BUSE
- Keywords
- Youth Mortality
- Hybrid Lee-Carter Models
- Supervisor
- Mr. E. Mukonoweshuro
