FOREIGN EXCHANGE PRICE PREDICTION USING MACHINE LEARNING

• Main Author: Eko Trinowo, Luthfi
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56133
• Institution: Institut Teknologi Bandung

The currency exchange rate is one of the important factors to measure the success of a country. Failed to plan their steps means that their currency will become worthless. Predicting the movement of the foreign exchange rate is important, considering many relations between countries that require money exchange between two of them. Whether we talk about import, export, or even a country's debt, all financial activities that need foreign exchange rates have to think carefully. Knowing when a currency should be exchanged is important to keep their money rate in the foreign exchange market from being affected by massive inflation. However, there are many parameters that need to be considered when determining the future value of foreign exchange. With today's advanced technology, prediction on historical data with time-series form is not more impossible. In this study, machine learning technology, both shallow and deep learning, used to predict the price of foreign exchange. By focusing on five foreign exchange such as USDIDR, EURIDR, GBPIDR, CHFIDR, and also JPYIDR, and by using USDIDR as the main reference in conducting the experiments to obtain the optimum parameters, four different machine learning models were used to evaluate which model had the best performance to predict the price of foreign exchange. Model's performance evaluated using MAPE metric and it was found that XGBoost has the best performance between another model. XGBoost had an average MAPE score of only 0.621%, followed by random forest with 0.682%, then LSTM with 1.005%, and GRU with 1.134%. This value obtained by using a set of parameters that has the most optimal MAPE in each model.