Ulfa Siti Nuraini, Dinda Galuh Guminta, Ibnu Febry Kurniawan, Agnes Ona Bliti Puka
This paper compares the statistical method Seasonal Autoregressive Integrated Moving Average with Exogenous variables (SARIMAX) and the machine learning method Long Short-Term Memory (LSTM) for modelling sales of transactional data. The main contribution of this work is the effect of the sum of user and twin dates as an exogenous variable and the seasonal pattern in modelling e-commerce transactional data. To validate the accuracy of the model, various scenarios are employed, including seasonal parameters, AR and MA parameters for SARIMAX, and the number of layers and nodes for LSTM. It is found that the forecasting SARIMAX model in December 2023 yields a lower RMSE (43.28) than the LSTM model, as indicated by the forecasting pattern that closely follows the real data in the testing dataset. This research is highly relevant as it provides a practical forecasting tool for transactional data that uses promotions in special events. ©2025 IEEE.
Department of Data Science, Universitas Negeri Surabaya, Surabaya, 60231, Indonesia; Mathematics Education Department, Larantuka Institute of Teacher Training and Technology, 86212, Indonesia
