A Novel FIG-LSTM Ensemble Machine Learning Technique for Currency Exchange Rate Forecasting

Accurately predicting currency exchange rate behaviour remains a major challenge for all stakeholders (e.g. traders, investment firms, banks, etc.) in the foreign exchange (forex) market. Developing machine learning models that offer
more accurate and potentially more reliable predictions is identified as a critical objective for the forex market. To address this issue, this paper proposes an ensemble machine learning model that integrates fuzzy information granule (FIG) with long short-term memory (LSTM) in a gated recurrent unit (GRU) to achieve a better forex forecasting performance. The proposed model uses open, high, low, close (OHLC) data and relevant technical indicators such as moving average, bollinger bands, %b, bandwidth, moving average convergence divergence (MACD), relative strength index (RSI), and average true range (ATR) as inputs. The outputs of the combined FIG and LSTM models are passed into a trained GRU model to make the final forex prediction. To evaluate the predictive performance of the proposed model, experiments are conducted using one-day candles of three of the most traded currency pairs, EUR/USD, USD/GBP and USD/CAD from 01 August 2019 to 31 December 2023 data set. The proposed model shows better forecasting performance in terms of root mean squared error (RMSE), mean
absolute percentage error (MAPE), mean absolute error (MAE), and coefficient of determination (R2) values when compared with conventional LSTM, FIG and GRU prediction models. The proposed FIG-LSTM model also outperforms a state-of-the-art GRU-LSTM hybrid prediction model.