A Hybrid Model Based on Multi-LSTM and ARIMA for Time Series Forcasting

In recent years, deep learning has rapidly developed and been widely applied across different fields. While statistical models are known for their powerful interpretability and unique processing mechanisms, time series are a common form of data that have found widespread use in areas such as economics, finance, and environmental science. The primary goal of time series analysis is to reveal the laws and characteristics of the data to make accurate predictions and informed decisions. However, since different methods have their own advantages and disadvantages, hybrid models can effectively combine them to expand the benefits and reduce the drawbacks. This paper proposes a time series forecasting model based on a multivariate LSTM with ARIMA. The ARIMA method is first used to analyze the time series data and obtain prediction results and confidence intervals. Then, these results are combined with other multivariate variables to serve as input features for the multivariate LSTM model to predict and model the time series data. The study selects the opening, closing, high, and low prices of Apple stock in the United States from March 1, 2022, to March 1, 2023, for prediction data. And the prices from March 1, 2023 to March 15, 2023 are for the following test. The results demonstrate that the proposed method is more effective than traditional methods.