Dynamic Model Selection Based on Demand Pattern Classification in Retail Sales Forecasting

Many forecasting techniques have been applied to sales forecasts in the retail industry. However, no one prediction model is applicable all cases. For demand of same item, different results models often confuse retailers. large companies with a wide variety products, it difficult find suitable for each item. This study aims propose dynamic selection approach that combines individual and combination based on both patterns out-of-sample performance Firstly, metrics squared coefficient variation (CV2) average inter-demand interval (ADI), we divide items into four types: smooth, intermittent, erratic, lumpy. Secondly, select nine classical methods M-Competitions build pool models. Thirdly, design two weighting strategies determine final prediction, namely DWS-A DWS-B. Finally, verify effectiveness this by using datasets from an offline retailer online China. The empirical show these can effectively improve accuracy forecasting. method intermittent lumpy, while DWS-B smooth erratic.