Research On Prediction Of Exchange Rate Based On Machine Learning And Combination Model

Exchange rate is an important macroeconomic variable,which affects the stability of national economy and global financial system.Therefore,the effective prediction of exchange rate has important practical significance.Based on this,this paper provides a new exchange rate forecast method based on time series model and machine learning model.This paper uses combination models to predict the exchange rate based on the exchange rate itself.Due to the high noise characteristics of exchange rate data,this paper first uses singular spectrum analysis to denoise the data,and then optimizes the combined prediction model of the residual method and the combined prediction model of the weight method.1.Optimization of combined model based on residual method.The residual method uses ARIMA model to extract the linear subject of exchange rate data,then uses machine learning to predict the residual part,and the final predicted value is the sum of the two parts.At present,the prediction of residuals in the field of exchange rate forecasting is rarely studied using deep learning.Therefore,this paper makes further research on the basis of the existing research,that is,using the ARIMA model to predict the linear main part of the exchange rate data,using the deep PSO-LSTM model to predict the residual part,and finally build the SSA-ARIMA-PSO-LSTM combination prediction model,and the empirical results show that the combined model predicts well.2.Optimization of combination model based on weighting method.Single models are the key to the combined model.The selection of single models should take full account of the prediction performance of the model and the correlation among the models.Models with less correlation can play their own diferent advantages and achieve complementary advantages.In this paper,the Pearson correlation coefficient is used to analyze the correlation of the prediction results of the four single models.Then,according to the results of correlation analysis,we combine the ARIMA,XGBoost,PSO-SVR and PSO-LSTM models.Finally,we build a new combination forecasting model based on information entropy,and the combined prediction model is compared with three other combined models,which further confirms the superiority of the combined prediction model based on information entropy proposed in this paper.3.Combined with RMSE,MAE and MAPE,the prediction effect of the combined model based on information entropy constructed in this paper is better than that of the residual method combined model.Therefore,the innovations of this paper are as follows:1.When using the residual method to predict the exchange rate,the LSTM model is used to predict the residuals of exchange rate data for the first time,and the prediction effect is good.2.The entropy combination method is introduced into the field of exchange rate prediction,and the combined prediction model of SSA-ARIMA-XGBoost-SVR-LSTM is constructed.The results show that the prediction effect of the combined prediction model is better than the four single models and the other three combinatorial models.