Weather sensitive demand forecasting method based on SVR for shoes products

Weather Sensitive Demand is defined as abnormal variation of demand from seasonal fluctuation because of weather condition's abnormal fluctuation. The majority of retailers acknowledge the impacts of weather. However, none of the conventional predictive modeling processes adequately address the impact of weather. In this paper, a weather sensitive demand forecasting method based on support vector machine (SVM) is proposed, in which the weather is taken as a very important impact factor for shoes & apparels retailers. Firstly, weather sensitive transformer is developed to transform the temperature factor to Heating Degree Days (HDD) and Cooling Degree Days (CDD), and then the most relative factors are selected from the other weather factors, such as the rainfall and the humidity by using Recursive Feature Elimination (RFE) based on SVM. Secondly, Particle Swarm Optimization (PSO) is employed to optimize the parameters of SVM to acquire demand forecasting model with better performance. Finally, real-world evaluation on a Chinese shoes & apparels retailer shows that the effectiveness of the proposed method.