The comparison of forecasting analysis based on the ARIMA-LSTM hybrid models

Abstract

For ARIMA-LSTM hybrid model, it can be mixed in different ways. Because ARIMA model can be separated into three parts: trend, seasonality, and residuals. One way is that using LSTM model to forecast residuals of ARIMA model, while another way is that using LSTM model to forecast seasonality and residuals of ARIMA model, RMSE, MAE, MAPE are used to measure the fit of two different ARIMA-LSTM hybrid models with the data of 2010–2020 Shenzhen maximum temperature every 3 hours. In result, it is concluded that using LSTM model to forecast residuals of ARIMA model is a better way to forecast than another

Identifiers

DOI

10.1109/icecem54757.2021.00044

Citations and references