A Selection of Advanced Technologies for Demand Forecasting in the Retail Industry

Retail companies always attempt to find a forecasting method to balance their purchasing and sales, whereas the performances of various prediction techniques are still not reliable. Furthermore, it is a question that how to select the proper forecasting model for some specific type of products. In this research, the classical forecasting models and the latest developing forecasting technologies are compared together based on the perishable products and non-perishable items respectively. The process is designed to compare the performance of typical statistic methods with several machine learning methods based on the thousands of historical transaction record of a large grocery retailer. The criterion is also explored in this study include predictive performance, generalization ability, runtime, cost and convenience to evaluate the comprehensive performance of these models, thus companies can easily choose their most accepted model.