Intermittent Demand Forecasting Using GRU Models: Focusing on the Volatility and Irregularity of Imported Car Demand in South Korea

This study investigates demand forecasting for imported cars using the GRU (Gated Recurrent Unit) model. To enhance the forecasting accuracy for the intermittent demand of imported cars, the sales time series data for different car models are categorized into four groups based on demand size volatility (ADI) and the irregularity of zero demand (CV2). The GRU model is then applied to each group, and the forecasting performance of each group is compared. To further improve the model&quots forecasting accuracy, the study adjusts the data sample size, input window size, and output window size, proposing optimized data input/output window sizes for each group. Two main analyses are conducted: First, whether the forecasting performance shows the same pattern across groups based on demand size volatility and zero demand irregularity is examined. Second, the impact of data sample size and input/output window sizes on each group is individually assessed. The results indicate that using a 10-year data sample size provides the most stable forecasting performance across all groups. Additionally, as the output window size increases, the model&quots forecasting performance tends to decrease in each group. Conversely, the input window size exhibits a U-shaped relationship with model accuracy across all groups, with the inflection point varying according to data size volatility.