Forecasting e-Journal Unique Visitors using Smoothed Long Short-Term Memory

Abstract

The number of unique visitors every day can be used as a benchmark to assess the success of an electronic journal. Unique visitors are visitors who use one IP in a certain period. With an increasing number of unique visitors every day, it can be inferred that scientific periodicals are increasingly in demand by the wider community, affecting the breadth of distribution and speeding up the journal accreditation system. Therefore, a model that can predict the number of unique visitors in electronic journals in the future can be beneficial for journal administrators. This paper evaluates Long Short-Term Memory (LSTM) performance when predicting the next day’s (T+1) time series for the number of unique visitors to an e-Journal website. The unique visitor data sample used was from January 1, 2018, to December 31, 2018. The distribution of data testing and training used was 80% and 20%. The data quality has been improved by smoothing the input data using exponential, mean, and median smoothing. The evaluation results show that the architecture with the best performance is model 2, namely the model mean + LSTM architecture 1-5-1 with a learning rate of 0.2 and a MAPE value of 0.08098. Therefore, we conclude that data smoothing with mean smoothing can improve Long Short-Term Memory performance for unique website visitor forecasting.