EVALUATION OF TIME-SERIES MODELS FOR PREDICTING STATUTORY NOTIFIABLE INFECTIOUS DISEASES IN SHANDONG PROVINCE, PR CHINA

Abstract

In Shandong Province, PR China, three time-series models were utilized to forecast and analyze statutory notifiable infectious diseases.  The models’ predictive performances also were assessed.  Holt-Winters, Prophet and Seasonal AutoRegressive Integrated Moving Average (SARIMA) models were trained using a dataset of infectious diseases’ incidents that was gathered between January 2017 and December 2021.  The test set consisted of time-series data on incidents from January to December 2022, and the fitting effects were assessed.  The monthly incidences of infectious diseases were predicted using the three time-series models, and the best-performing model was chosen based on four different error indicators: mean squared, mean absolute, mean absolute percentage, and symmetric mean absolute percentage errors.  The Prophet model, followed by the Holt-Winters and SARIMA models, produced for all indicators the best goodness of fit for acute hemorrhagic conjunctivitis, gonorrhea, mumps, syphilis, tuberculosis, and viral hepatitis.  For acquired immunodeficiency syndrome, acute hemorrhagic conjunctivitis, epidemic hemorrhagic fever, gonorrhea, mumps, syphilis, tuberculosis, and viral hepatitis, the Prophet model fared better than the SARIMA model; while for acquired immunodeficiency syndrome, acute hemorrhagic conjunctivitis, brucellosis, epidemic hemorrhagic fever, gonorrhea, mumps, other infectious diarrheal diseases, syphilis, tuberculosis, and viral hepatitis, the Prophet model performed better than the Holt-Winters model.  The Prophet model fits curves more accurately by incorporating seasonal and cyclical variations.  Taken altogether, the Prophet model outperforms both the Holt-Winters and SARIMA models in terms of yielding more accurate trend predictions regarding disease outbreaks.  This should help public health agencies formulate their prevention and control programs with more precise results.