A Comparative Analysis of NNAR and LSTM Models for Short-Term COVID-19 Forecasting in Saudi Arabia

The COVID-19 pandemic has posed an ongoing challenge for public health systems around the globe. Accurate forecasting of daily confirmed COVID-19 cases in Saudi Arabia has remained critical for informed planning and timely interventions. This research explores and compares the predictive performance of two artificial neural network models—Nonlinear Autoregressive Neural Network (NNAR) and Long Short-Term Memory (LSTM)—applied to Saudi Arabia’s COVID-19 case data from March 2020 through December 2021. Using standard evaluation metrics, including MAE, RMSE, MAPE, and Theil’s U, the study demonstrates that the NNAR model provides slightly more stable and accurate predictions in short-term horizons than LSTM. While LSTM models are known for capturing complex temporal patterns, our findings suggest that NNAR may offer a more robust option in volatile epidemiological conditions. These insights contribute to the growing field of epidemic forecasting and provide practical considerations for health policymakers in the region.