A Comparison between ARIMA, LSTM, and GRU for Time Series Forecasting

By P. T. Yamak, Li Yujian, P. K. Gadosey. International Conference on Advances in Computing and Artificial Intelligence, 2019

381 citations recorded by the source corpus.

A critical area of machine learning is Time Series forecasting, as various forecasting problems contain a time component. A series of observations taken chronologically in time is known as a Time Series. In this research, however, we aim to compare three different machine learning models in making a time series forecast. We are going to use the Bitcoin's price dataset as our time series data set and make predictions accordingly. The results show that the ARIMA model gave better results than the deep learning-based regression models. ARIMA gives the best results at 2.76% and 302.53 for MAPE and RMSE respectively. The Gated Recurrent Unit (GRU) however performed better than the Long Short-term Memory (LSTM), with 3.97% and 381.34 of MAPE and RMSE respectively.

Subfields: Artificial Intelligence, Management Science and Operations Research, Signal Processing

View Paper: https://www.semanticscholar.org/paper/5df5df19deb61524c4a2003e716c80caf22f47f3

Citations in the corpus

• A Review of ARIMA vs. Machine Learning Approaches for Time Series Forecasting in Data Driven Networks (2023, neutral): “ARIMA methods prevail over deep learning algorithms in bitcoin price prediction applications in the works of Nguyen et al. [44] and Yamak et al. [6], and this result may be due to several factors, as the chosen values of the model parameters and the total amount of data can affect the results of…”