Gated Recurrent Unit (GRU) Method in Predicting Stock Prices of PT Astra Agro Lestari Tbk. on the Indonesia Stock Exchange (IDX)

Abstract

Gated Recurrent Unit (GRU) is one of Recurrent Neural Network (RNN) algorithm architecture that is considered effective in processing sequential data and processing high-frequency data such as stock data. The study was conducted to assess the performance of the GRU model in forecasting daily stock prices and short-term predictions for 30 days. Hyperparameter tuning is applied to optimize the number of hidden layer units, dense layers, batch size, and also droplets to improve accuracy and prevent overfitting. From the results of the study, the best model was built using 64 units of dense layer, 16 units of hidden layer, 16 batch size, and 0.1 dropout with an RMSE value of 0.013 and MAPE of 0.03 where the accuracy reached 99.97%. It can be concluded that the model shows good performance in predicting stock prices based on the evaluation of the model and forecasting results that are not too far from the latest actual data.

Keywords: GRU, Machine Learning, MAPE, RMSE, Stock, Time Series.