Time Series Forecasting Based on ARIMA and LSTM Models

Prediction is to speculate and judge the future development trend through scientific methods according to the past and present information of the predicted object. Nowadays, prediction models are applied in various fields which made them have great research value in the academic field. In this paper, two different sets of time series data are used for forecasting research. Through preliminary analysis, we find that these two groups of data have the characteristics of large sample size and nonlinearity. After studying the relevant literature, we find that the LSTM model seems to have more advantages than the ARIMA model when studying data with large sample size and nonlinear characteristics. To verify this finding, we forecast our two sets of data separately with these two models, and then compare the prediction accuracy of the two models by correlation coefficient test, and find that the LSTM model has better results in predicting these two sets of data. This provides some theoretical support for our conjecture.