LSTM Deep Learning vs ARIMA Algorithms for Univariate Time Series Forecasting: A case study

Abstract

This manuscript aims to study and compare the Long Short-Term Memory (LSTM) Deep learning to Auto regressive Integrated Moving Average (ARIMA) algorithms for a univariate time series, especially for stock price series. Using the mean absolute percentage error, the mean absolute error, or either root-mean-square deviation and according to our extracted dataset, we find that the classical approaches like ARIMA out-perform deep learning ones since they are very simple to use especially for linear univariate datasets. More specifically, LSTM deep learning algorithms are more powerful and provide better results in terms of predictions.