Sales Forecast for O2O Services - Based on Incremental Random Forest Method

This paper proposes an incremental random forest method to forecast the sales for the O2O take-out business. The proposed method has two characteristics. First, we identify the important features that contribute most to the forecast accuracy by deleting the noisy features. This feature selection process helps to improve the forecast accuracy. Second, we use an incremental method based on random forest by adding incremental features and focus on sales increment prediction. This incremental random forest method could further help to control the forecast error. Moreover, we apply a real data set from an online merchant in one of the largest O2O platforms to validate our method. The results show that the feature selection can significantly reduce the mean absolute percentage error (MAPE) by 11.64%. Furthermore, the incremental random forest method further reduces the MAPE by 3.10%. With a better sales forecast, the merchant can improve its operations for replenishment and marketing.