Forecasting Tuberculosis Incidence in Somalia: A Comparative Analysis of Single and Hybrid Time-Series Models

Abstract Tuberculosis (TB) remains a significant public health challenge, necessitating accurate forecasting methodologies for effective control and prevention strategies. This paper explores the application of hybrid models for forecasting TB incidence in Somalia. The study employs a comprehensive suite of 14-time series models, including five single models—ARIMA (Autoregressive Integrated Moving Average), ETS (Error Trend Seasonality), TBATS (Trigonometric seasonality, Box-Cox transformation, ARMA errors, Trend, and Seasonal), Theta, and NNAR (Neural Network Auto Regression)—and nine hybrid models (ARIMA-ETS, ARIMA-NNAR, ARIMA-Theta, ARIMA-TBATS, ARIMA-ETS-Theta, ARIMA-ETS-NNAR, ARIMA-ETS-TBATS, ARIMA-Theta-NNAR, and ARIMA-TBATS-NNAR). Annual TB incidence data from 2000 to 2022, sourced from the World Bank, is utilized to train and evaluate the models. Model performance is assessed using metrics such as Theil's U statistic, Mean Absolute Percentage Error (MAPE), Symmetric Mean Absolute Percentage Error (SMAPE), and Root Mean Square Error (RMSE). The TBATS model demonstrates the best fit among the single time series models, while the ARIMA-ETS-TBATS hybrid model outperforms the other hybrid models. The resulting forecasts provide valuable insights into the future TB incidence trends in Somalia, aiding in informed public health decision-making and targeted intervention strategies. The study underscores the importance of hybrid modeling for enhanced forecasting accuracy in the context of TB control efforts.