A Study on the Prediction Method Based on ARIMA-LSTM Combined Model in Wordle Game

Abstract

In this paper, the paper propose a new integrated prediction model AMIRA-LSTM model, which integrates the advantages of each of the two models, firstly, the paper use ARIMA and LSTM models to fit for the known time series data of the wordle game respectively, and find that these two models fit better for the linear and nonlinear parts of the samples respectively, and in this paper, the paper substitute the ARIMA( p,d,q) model, which portrays the linear part of the data series, is substituted into the LSTM model to obtain the prediction about the residuals, thus modifying the prediction of the ARIMA model, and thus organically combining with the respective advantages of the LSTM model describing the nonlinear part of the data series, the final sample prediction results almost overlap with the actual results, and the RMSE and MAPE are 0.0256 and 0.33, respectively; this paper determines the word The RMSE and R2 are 0.2993 and 0.9021, respectively, with high model accuracy.