Prediksi Harga Xau/USD Menggunakan Metode Bidirectional Long Shot-Term Memory (Bilstm)

Abstract

The price movement of gold in the form of the XAU/USD currency pair has high volatility because it is influenced by global gold price dynamics and fluctuations in the US dollar exchange rate. These conditions cause a high level of uncertainty and make it difficult for traders and investors to make optimal transaction decisions. Therefore, a prediction method is needed that is capable of accurately capturing price movement patterns based on historical data. This study aims to apply the Bidirectional Long Short-Term Memory (BiLSTM) method in predicting the closing price of XAU/USD on a daily time frame and evaluating the accuracy of the prediction results. The data used in this study is historical XAU/USD data from January 2020 to January 2026 obtained from the MetaTrader 5 platform. The variables used include Open, High, Low, Close, and Tick Volume. The BiLSTM model was built with one bidirectional layer and two dense layers, and automatic hyperparameter tuning was performed to obtain the best configuration. The model was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). The results indicate that the best configuration was achieved at time step 45, producing a testing RMSE of 49.09 and a validation loss of 0.001986. The 30-day forecasting evaluation shows that the model maintains optimal accuracy for up to five consecutive days with a MAPE value of ≤ 2%. These findings demonstrate that the BiLSTM method has strong capability in short-term XAU/USD price prediction and potential application as a decision support system in trading and data-driven market analysis.