A Review of Long Short-Term Memory Approach for Time Series Analysis and Forecasting

Abstract

AbstractThe long short-term memory (LSTM) approach has evolved into cutting-edge machine learning techniques. It belongs to the category of deep learning algorithms originating from Deep Recurrent Neural Network (DRNN) forms. In recent years, time series analysis and forecasting utilizing LSTM can be found in various domains, including finance, supply and demand forecasting, and health monitoring. This paper aims to analyze the previous recent studies from 2017 to 2021, emphasizing the LSTM approach to time series analysis and forecasting, highlighting the current enhancement methods in LSTM. It is found that the applications of LSTM in the current research related to time series involve forecasting or both. The finding also demonstrated the current application and advancement of LSTM using different enhancement techniques such as hyperparameter optimization, hybrid and ensemble. However, most researchers opt to hybridize LSTM with other algorithms. Further studying could be applied to improve LSTM performance, especially in the domain study, in which the LSTM enhancement technique has not been widely applied yet.KeywordsLong short-term memoryTime series analysisTime series forecastingDeep learning