Comparison of RNN, LSTM, and GRU Methods on Forecasting Website Visitors

Abstract

Forecasting is the best way to find out the number of website visitors. However, many researchers cannot determine which method is best used to solve the problem of forecasting website visitors. Several methods have been used in forecasting research. One of the best today is using deep learning methods. This study discusses forecasting website visitors using deep learning in one family, namely the RNN, LSTM, and GRU methods. The comparison made by these three methods can be used to get the best results in the field of forecasting. This study used two types of data: First Time Visits and Unique Visits. The test was carried out with epoch parameters starting from 1 to 500 at layers 1, 3, and 5. The test used first-time visit data and unique visit data. Although tested with different data, the test results obtained that the smallest MSE value is the LSTM method. The value of each MSE is 0.0125 for first-time visit data and 0.0265 for unique visit data. The contribution of this research has succeeded in showing the best performance of the three recurrent network methods with different MSE values.