Website-Based Gold Price Movement Prediction System Using the Long Short-Term Memory (LSTM) Method

Gold price prediction is an important aspect in supporting investment decision making amid dynamic market fluctuations. This research aims to find the best hyperparameter combination in building a gold price prediction model using the Long Short-Term Memory (LSTM) method through the Grid Search and Bayesian Optimization tuning approaches. The data used is historical gold price data from Yahoo Finance for the last 10 years (2015-2025) which includes date attributes, opening price, closing price, highest price, lowest price, and trading volume. This study was conducted with two data divisions, namely the 70%:30% and 80:20 ratios, to evaluate the performance of the model to find optimal results. The hyperparameter tuning process includes finding optimal values for epoch, batch size, learning rate, number of neurons, dropout, and optimizer parameters. Model evaluation was conducted using MAE, RMSE, and MAPE metrics and the best results were obtained from tuning using Grid Search at a split ratio of 70%:30% with MAE values of 19.5470, RMSE of 26.5331, and MAPE of 0.93%. The system automatically updates the daily model by using web scraping technique and utilizing Python scheduler. The system was developed based on a website using the Flask framework and has an interactive display consisting of dashboard, historical, and prediction pages. The test results show that the system runs according to its function and the LSTM model is able to predict gold prices with good accuracy. This research shows that proper hyperparameter tuning can significantly improve the performance of the prediction model.