Time Series Forecasting of E- Databases Subscription Mahidol University Library with Exponential Smoothing, LSTM, and ARIMA Models
Issued Date
2020
Resource Type
Language
eng
Rights
Mahidol University
Rights Holder(s)
IEEE Explore
Suggested Citation
Vanaphol Chamsukhee (2020). Time Series Forecasting of E- Databases Subscription Mahidol University Library with Exponential Smoothing, LSTM, and ARIMA Models. doi:10.1109/InCIT50588.2020.9310785 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/62005
Title
Time Series Forecasting of E- Databases Subscription Mahidol University Library with Exponential Smoothing, LSTM, and ARIMA Models
Author(s)
Other Contributor(s)
Abstract
To forecast the number of access to E-Databases website is very important electronic databases subscript planning, electronic databases renewal, maintenance time, performance systems, etc. The EZproxy server is generated monthly log files by collecting the user activities access to e-database website. This paper uses the internet access log in year 2019 using three forecasting models in order to compare and identify the most appropriate model for forecasting the number of access to the website in the future. The comparison are made by evaluating model with Mean Square Error (MSE) method on three models which are ETS, LSTM, and ARIMA. The MSE results for each model are 0.150, 0.127, and 0.153 respectively. The LSTM model is the best model to obtain the minimum average error value and has shown suitability with such as time series data and seasonality including number of access to E-Database can be precise training model.
Description
5th International Conference on Information Technology (InCIT), 21-22 Oct. 2020, Chonburi, Thailand. p. 202 - 207