Adaptive Holt-Winters Forecasting Method based on Artificial Intelligence Techniques

The fundamental decision-making of a firm is forecasting. It is an important activity that affects the performance of a company. Among forecasting tools, exponential smoothing techniques are the most relevant in industries. They yield exceptional results with low forecasting errors. The triple exponential smoothing technique, viz. the Holt-Winter (HW) method, is the most popular when the seasonality is embedded in the data. However, the three smoothing parameters predetermined by the analyst are still problematic in practice. We proposed two improved HW methods in this study by combining two artificial intelligence techniques to adapt the three smoothing parameters iteratively. The proposed methods are tested by forecasting a local stainless steel price data set. We found that the PSO-HW method outperforms the traditional HW and GSA-HW method in the mean absolute percentage error measurement. However, the GSA-HW method surpasses the other two methods in the direction accuracy percentage.