Optimalisasi Prediksi Kasus COVID-19 di Indonesia: Perbandingan Teknik Validasi 80-20 Split dan Walk-Forward dengan ARIMA

Asrul, Divanda Arya Inasta
Soebroto, Arief Andy

J-Intech (Journal of Information and Technology)
Vol 12 No 02 (2024): J-Intech : Journal of Information and Technology

This study presents a comparative analysis of the 80-20 split and walk-forward validation techniques for forecasting daily COVID-19 cases in Indonesia using the ARIMA model. Building on previous research, the ARIMA model has proven effective in various epidemiological contexts; however, this study highlights the critical importance of selecting the appropriate validation technique. The study uses data from January 3, 2020, to October 18, 2023, to develop a predictive model evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The findings indicate that the walk-forward validation technique outperforms the 80-20 split, with MAE of 137.32 and RMSE of 198.23, compared to the 80-20 split MAE of 4190.92 and RMSE of 4479.15. These results suggest that walk-forward validation provides more accurate and reliable predictions, particularly for dynamic and non-stationary data scenarios. This study underscores the significant impact of validation technique selection on ARIMA model performance, contributing new insights into forecasting methodologies in epidemiology.