Application of Vanilla Long Short-Term Memory Networks (LSTM) and Auto-Regressive Integrated Moving Average (ARIMA) on exchange rate forecasting / Mysarah Haslan and Nor Hayati Shafii

Predicting foreign exchange rates is a difficult task in the area of financial forecasting. Changes in exchange rate affected the country’s rate of economic growth. There are a lot of forecasting models used in order to predict the future value of the exchange rate. This study aims to determine the most accurate model between two different machine learning models which are Vanilla Long-Short Term Memory (LSTM) and Auto-Regressive Integrated Moving Average (ARIMA) in predicting the exchange rate of Malaysian Ringgit (MYR) and United State Dollar (USD). In addition, this study used a statistical package in Python software that uses machine learning to better handle the challenge of time series forecasting. Vanilla LSTM and ARIMA are trained using Python software in order to train the dataset. Coding programming in Python software runs to make better analysis to achieve the accurate model. Prediction is also made after the comparison of error measures of two models. The result of the comparison between the two models showed that the MSE and RMSE of the Vanilla LSTM is lower than the ARIMA model. The Vanilla LSTM model overcomes the ARIMA in forecasting the exchange rate. Therefore, the analysis of the study obtained that the vanilla LSTM model is the most accurate model to make predictions on the exchange rate with 0.0102 and 0.1011 for MSE and RMSE respectively. While for the ARIMA with 0.0113 and 0.1062 of MSE and RMSE respectively. The final prediction for July 2022 is RM 4.22.