Título: Forecasting Mortality Rates: A Comparative Analysis of LSTM -Based and Lee – Carter Models
Autores: Caroliny Rodrigues Nascimento, Roberta A de A Fagundes
Resumo: Accurate mortality forecasting is essential for informing long-term planning in areas such as public health, pension systems, and insurance. However, capturing the underlying patterns in mortality data remains challenging due to demographic shifts, regional disparities, and the presence of complex nonlinear trends over time. This study investigates the application of deep learning architectures (Long Short-Term Memory, Bidirectional Long Short-Term Memory, and Convolutional Neural Networks combined with Long Short-Term Memory) for forecasting mortality rates, and compares their predictive performance with the traditional stochastic Lee-Carter model. Mortality data from the Human Mortality Database, covering the United States from 1941 to 2020 and segmented by demographic region and gender, were used to train and evaluate the models. A hyperparameter optimization process using random search was conducted prior to model training. Performance was assessed using mean absolute error and root mean square error. The results demonstrate the superior performance of deep learning models compared to the Lee-Carter model. These findings support the use of neural networks as a robust and flexible alternative for mortality forecasting in demographic studies.
Palavras-chave: Mortality rate; Machine Learning; LSTM; Bi; LSTM; CNN-LSTM.
Páginas: 6
Código DOI: 10.21528/CBIC2025-1168761
Artigo em PDF: CBIC_2025_paper1168761.pdf
Arquivo BibTeX:
CBIC_2025_1168761.bib