Prediction of Stock Prices Using a Combination of LSTM and GRU Algorithms on LQ45 Indexed Stocks

Stock investment is a popular choice due to its high profit potential, but stock price movements influenced by various factors make predictions challenging. This study combines LSTM and GRU algorithms in an LSTM-GRU-LSTM-GRU configuration to predict the stock prices of five companies from the LQ45 index (ADRO, ANTM, ESSA, PGAS, PTBA) using daily closing prices and transaction volumes. The model’s performance was evaluated using different dataset splits (70:30, 80:20, 90:10) and epoch parameters (10, 20, 30, 40, 50), measured using RMSE and MAPE metrics. The results show that combining LSTM and GRU models provides reasonably accurate predictions with specific dataset compositions and epoch settings. For ADRO, the best combination was 90:10 with 30 epochs (RMSE 0.0183, MAPE 2.3619%); for ANTM, 90:10 with 30 epochs (RMSE 0.0110, MAPE 1.5968%); for ESSA, 80:20 with 40 epochs (RMSE 0.0103, MAPE 3.7750%); for PGAS, 90:10 with 40 epochs (RMSE 0.0046, MAPE 2.5603%); and for PTBA, 70:30 with 50 epochs (RMSE 0.0066, MAPE 2.7489%). The model shows a good ability to capture stock price trends, despite some significant deviations, particularly in PTBA and ADRO stocks. These findings affirm that the combination of LSTM and GRU can enhance stock price prediction accuracy, with further improvements possible through model parameter adjustments to reduce deviations.