PREDIKSI HARGA EMAS MENGGUNAKAN METODE TIME SERIES LONG SHORT - TERM MEMORY NEURAL NETWORK

Gold is one of the popular investment tools among people who are resistant to inflation. However, gold investment activities have time series data type risks. So that people need to have knowledge as a guide when carrying out gold investment activities, namely by predicting the future price of gold to minimize risk. Long Short-Term Memory is a derivative of the RNN method which can be used in predicting time series data. This study aims to predict the price of gold from the 1-day time series data that has been collected from the harga-emas.org website to determine the value of prediction errors using the LSTM method. The parameter analysis carried out in this study was the number of hidden neurons, the learning rate, and the epoch. The best combination of parameters produced in this study were 16 hidden neurons, a learning rate of 0.01, and 100 epochs. The best values produced in this study were RMSE 9139.14318 and MAPE 0,69794%. The best MAPE error calculation in this study is the research "Gold Price Prediction Using Feed Forward Neural Network with the Extreme Learning Machine Method" which produces the best MAPE 0,8065%. These results indicate that the MAPE error in this study is better than that research and the model formed can be said to be very good because the best MAPE value generated is below 10%