PREDIKSI HARGA EMAS BERBASIS LONG SHORT TERM MEMORY (LSTM) UNTUK KEPUTUSAN INVESTASI

Gold price movements are an important indicator in making investment decisions, especially in dynamic and uncertain markets. This research aims to develop a gold price prediction model based on Long Short-Term Memory (LSTM), a type of Recurrent Neural Network (RNN), which is able to capture temporal patterns in historical gold price data. This model is expected to provide more accurate predictions than conventional methods, so that it can be an aid in making investment decisions. The data used in this research includes daily gold prices over a certain period of time, combined with other economic indicators such as oil prices, currency exchange rates and interest rates. The data is processed through pre-processing, including normalization and dividing the data into
training, validation, and test sets. LSTM models are designed with hidden layers that are optimized to capture complex relationships between data. The research results show that
the LSTM model is able to predict gold prices with a high level of accuracy, demonstrated by evaluation metrics such as Mean Squared Error (MSE) and Mean Absolute Error (MAE). Compared to linear regression methods and ARIMA models, LSTM models
provide superior performance in capturing long-term trend patterns and short-term price fluctuations. In conclusion, The LSTM model used in this study demonstrated excellent predictive performance, with evaluation metrics showing an MAE of 26.12, MSE of 1269.15, RMSE of 35.63, and an R² value of 0.9858, indicating that the model can explain 98.58% of the variance in actual gold price data. The model architecture consists of a single LSTM layer with 64 neurons and one output layer, trained over 50 epochs with a batch size of 32, resulting in stable predictions without overfitting. Visualization results
show that the model effectively follows gold price trends and consistently predicts a steady price increase over the next 30 days. The implementation of the model into a web-based system using TensorFlow.js allows users to access real-time predictions efficiently and responsively.
Keywords: Gold price prediction, Long Short-Term Memory (LSTM), investment decisions, historical data, deep learning models.