Forecasting Regional Order Quantities in E-commerce Websites Using Time Series Models

The rapid adoption of electronic commerce (EC) services has led to the establishment of numerous online sales platforms. Forecasting the order quantity is crucial for effective inventory management at EC-affiliated stores and meeting demand in EC services. In this study, we explored time series modeling, focusing on ARIMA-derived models, namely SARIMA and SARIMAX, considering the limited features in the dataset for forecasting. The dataset was obtained from an EC platform specializing in floral products with peak demand during Japanese Mother’s Day. In the SARIMAX models, we proposed exogenous variables such as binary indicators for Mother’s Day and holidays and a variable denoting the week of May. The SARIMAX model with the “Mother’s Day” variable yielded the best performance. However, forecasting accuracy was inadequate due to date variability. To improve forecasting accuracy, we propose a data-formatting approach that expresses dates based on Mother’s Day. This approach aims to eliminate the influence of date variability. By adopting the proposed approach, we achieved more accurate forecasts compared to our previous results. In conclusion, our proposed exogenous variables and data-formatting approach allowed for order quantity forecasts with optimal accuracy. Despite the promising results, our study has limitations, such as the reliance on a specific dataset and need for further validation in diverse EC contexts. Future research could explore the integration of additional exogenous variables and investigate the scalability of our forecasting approach.