Application of Long Short-Term Memory (LSTM) Algorithm in Predicting Forex Trading Price Movements on the USD/JPY Pair

Miftahul Jannah (Universitas Malikussaleh)
Wahyu Fuadi (Department Of Informatics, Universitas Malikussaleh, Bukit Indah, Lhokseumawe, 24353, Indonesia)
Zahratul Fitri (Department Of Informatics, Universitas Malikussaleh, Bukit Indah, Lhokseumawe, 24353, Indonesia)

Proceedings of International Conference on Multidisciplinary Engineering (ICOMDEN)
Vol. 2 (2024): Proceedings of International Conference on Multidisciplinary Engineering (ICOMDEN)

The foreign exchange (Forex) market offers the potential for high profits but also great risks through currency pair trading. This research proposes the use of the Long Short-Term Memory (LSTM) algorithm, a type of Recurrent Neural Network (RNN), to predict forex price movements for the USD/JPY (American Dollar to Japanese Yen) currency pair based on daily data over a two-year period. The model was designed with a “sequential” architecture consisting of two LSTM layers with 100 units each, followed by a Dropout layer to reduce overfitting and a Dense layer to generate predictions. The total model has 365,905 parameters, with 121,301 parameters trained. During training, model evaluation showed that the combination of batch size 16 and epoch 150 resulted in an RMSE of 0.9840, indicating high accuracy. The application of the model also resulted in an RMSE value of 1.04 and a MAPE of 0.56%, with an average accuracy of 99.44%, indicating a prediction accuracy that successfully follows the actual price trend and is effective in capturing forex price movement patterns in the USD/JPY currency pair, thereby supporting future trading decisions.