Impact of Coiflet Wavelet Decomposition on Forecasting Accuracy: Shifts in ARIMA and Exponential Smoothing Performance

Accurate electricity demand forecasting is essential for efficient energy management and resource allocation. This study investigates the impact of Coiflet wavelet decomposition on the forecasting performance of Exponential Smoothing (ES) and ARIMA models. Two experimental approaches were employed: one using raw data and another incorporating wavelet denoising. Without wavelet transformation, ES performed better than ARIMA in the testing phase, with RMSE values of 13.62 and 13.93, MAE values of 11.22 and 11.54, and MAPE values of 3.04% and 3.14%, respectively. However, after applying wavelet decomposition, ARIMA showed significant improvement, reducing RMSE by 24.6%, MAE by 23.7%, and MAPE by 23.5% in the testing phase, outperforming ES. The hybrid ARIMA-wavelet model emerged as the most robust approach for forecasting electricity demand, demonstrating the effectiveness of wavelet-based denoising in improving predictive accuracy. These findings highlight the potential of integrating wavelet analysis with statistical forecasting models for more reliable time series predictions.