Studying Hybrid Models For Forecasting Foreign Exchange Rates By Using The Complete Ensemble Empirical Mode Decomposition With Adaptive Noise (CEEMDAN) With The Multilayer Long Short Term Memory (MLSTM)

Abstract

A study of a hybrid model for forecasting foreign exchange rates using Complete Ensemble Empirical Mode Decomposition With Adaptive Noise (CEEMDAN) in combination with a Multilayer Long Short Term Memory (MLSTM) schematic network model. that uses a filtering model to differentiate by frequency range for optimal data selection and using a short-term, long-term neural network model to predict the next data.

The purpose of this study was to study the efficiency and error in forecasting foreign exchange rate data of the hybrid model compare with the model without CEEMDAN data extraction and the sharing of the model between the same market data.

The data in the study consisted of the EURUSD GBPUSD JPYUSD AUDUSD exchange rates. Study the best MLSTM model from the daily EURUSD raw data using an Optimization method for forecasting in conjunction with the CEEMDAN data extraction to evaluate efficiency when forecasting the same market exchange rate data as other datasets. Using RMSE methods to compare the capabilities of a hybrid model using CEEMDAN with MLSTM and a model without CEEMDAN.

A study of the MLSTM model derived from optimization combined with CEEMDAN data extraction to predict the forward data. Found that for the EURUSD dataset ้has appropriate and represents the direction of the exchange rate movement, low tolerance, and the results were better than the not use CEEMDAN model. When using the model for other data sets on the market. The result is less efficient but is still highly effective for predicting exchange rates.