Evaluation of Time Series Forecasting Models for Predicting Lung Cancer Mortality Rates in the United States: A Comparison with Altuhaifa (2023) Study

This paper evaluates the performance of the following time series forecasting models - Simple Exponential Smoothing (SES), Holt's Double Exponential Smoothing (HDES), and Autoregressive Integrated Moving Average (ARIMA) - in predicting lung cancer mortality rates in the United States. It builds upon the work of Altuhaifa, which used Surveillance, Epidemiology, and End Results (SEER) data from 1975-2018 to evaluate these models. Altuhaifa's study found that ARIMA (0,2,2), SES with smoothing parameter $α=0.995$, and HDES with parameters $α=0.4$ and $β=0.9$ were the optimal models from their analysis, with HDES providing the lowest Root Mean Squared Error (RMSE) of 132.91. The paper extends the dataset to 2021 and re-evaluates the models. Using the same SEER data from 1975-2021, it identifies ARIMA (0,2,2), SES ($α=0.999$), and HDES ($α=0.5221$, $β=0.5219$) as the best-fitting models. Interestingly, ARIMA (0,2,2) and HDES yield the lowest RMSE of 2.56. To obtain forecasts with higher accuracy, an average model (HDES-ARIMA) consisting of HDES and ARIMA was constructed to leverage their strengths. The HDES-ARIMA model also achieves an RMSE of 2.56. The forecast from the average model suggests declining lung cancer mortality rates in the United States. The study highlights how expanding datasets and re-evaluating models can provide updated insights. It recommends further analysis using monthly data separated by gender, ethnicity, and state to understand lung cancer mortality dynamics in the United States. Overall, advanced time series methods like HDES and ARIMA show strong potential for accurately forecasting this major public health issue.