Abstract:

The accuracy of sales forecasting is crucial for e-commerce businesses to optimize inventory management, pricing decisions, marketing strategies and staff scheduling. At this point, different approaches such as statistical models, fuzzy systems, machine learning and deep learning algorithms are widely used for sales forecasting. This study investigates the performance of the deep learning based the Long-Short Term Memory (LSTM) model and the Facebook Prophet model on short-term sales forecasting. The performance of the proposed models is compared with the seasonal autoregressive integrated moving average (SARIMA) using real-life data from an e-commerce site. For the comparative analysis of the proposed forecasting models, weighted average absolute percent error (wMAPE), root mean square error (RMSE) and R-squared are selected as performance measures. The numerical results show that the LSTM model outperforms the Prophet and SARIMA models in terms of forecast accuracy for hourly sales forecasting.

Keywords: