A Clustering-based Sales Forecast Method for Big Promotion Days in O2O On-Demand Retailing

摘要

O2O on-demand service is a fast-growing business model which integrates traditional offline retailing and e-commerce. Big promotion is an effective way to attract new customers and promote sales in O2O service. However, high sales brought by big promotions can incur challenges for the operation management, especially for delivery capacity planning and replenishment scheduling. Using an accurate sales forecast is shown to be an effective way to alleviate such problems, though this technique still has its challenges. In this paper, we propose a clustering-based random forest (RF) method to predict the sale levels in shops for big promotion days. We further apply a dataset from one of the largest O2O grocery platforms in China to validate our method. The results show that the forecast accuracy of our proposed algorithm is 13.43 % higher than non-clustering RF method, 22.85% higher than non-clustering ARIMA method.