ARIMA-GRNN hybrid model

ARIMA-GRNN hybrid model forecasts infectious disease incidence, including hepatitis, by combining Autoregressive Integrated Moving Average (ARIMA) time-series modeling with a Generalized Regression Neural Network (GRNN) to capture linear and nonlinear patterns and improve predictive accuracy.

Key Features:

• Hybrid Approach: Integrates ARIMA and GRNN to capture both linear trends and nonlinear patterns in time-series disease incidence data.
• Seasonal and Trend Analysis: Uses ARIMA to model seasonal variations and trends, exemplified by application to hepatitis morbidity data from Heng County (January 2005–December 2012) showing a slight rising trend and clear seasonal variation.
• Neural Network Enhancement: Employs a non-parametric GRNN with an optimized smoothing factor to handle complex patterns without extensive parameter tuning.
• Performance Metrics: Evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Square Error (MSE), demonstrating superior performance relative to standalone ARIMA or GRNN models.

Scientific Applications:

• Public Health Decision Support: Provides forecasts of hepatitis incidence to inform resource allocation and timing of interventions for outbreak control.
• Epidemiological Forecasting: Applicable to forecasting other infectious diseases with seasonal and nonlinear incidence patterns using combined ARIMA and GRNN modeling.

Methodology:

Historical incidence data were collected from local CDCs (Heng County data: January 2005–December 2012); ARIMA(0,1,2)(1,1,1)12 was identified for the ARIMA component; a GRNN was developed with an optimized smoothing factor; ARIMA and GRNN were integrated into a hybrid framework; predictions were validated using data from January 2013 to December 2013.