Application of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) Algorithm in Gold Price Prediction

• Main Authors: Dewi, Ni Putu Juni Rahayu, Ginantra, Ni Luh Wiwik Sri Rahayu, Darma, I Wayan Agus Surya, Indrawan, I Gusti Agung
• Format: Conference Proceeding
• Language: English
• Subjects: Deep learning; Fluctuations; Gated Recurrent Unit (GRU); Gold; Gold Price Prediction; K-fold Cross Validation; Logic gates; Long Short-Term Memory (LSTM); Memory management; Supply and demand; Surveys
• ISSN: 2770-159X
• DOI: 10.1109/CITSM60085.2023.10455749

Gold is a precious metal that is in great demand by the public as an investment tool. The survey results stated that 58.5% of the 5.204 respondents chose gold as the most desirable investment. Gold has a fluctuating price. The price of gold can go up and down within a certain period of time. Fluctuations in world gold prices are influenced by economic factors such as supply and demand for gold. Therefore, we need a prediction that is able to estimate the price of gold based on the movement of gold prices in the previous period. This research compares the ability of the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) algorithms in predicting gold prices. This research is expected to obtain an algorithm with the best level of accuracy as a material consideration for investors in making investment decisions. The results of this research state that the Gated Recurrent Unit (GRU) algorithm is better than Long Short-Term Memory (LSTM) in predicting gold prices. The best GRU model uses batch size 8 and epoch 50 to obtain an RMSE of 1464.838, while the best LSTM model uses batch size 32 and epoch 50 to obtain an RMSE of 1469.144.