PREDICTION ANALYSIS OF INDONESIA STOCK PRICE INDEX USING GATED RECURRENT UNIT (GRU) AND LONG SHORT-TERM MEMORY (LSTM) METHODS
Econophysics is a field that we can use in predicting the value of stock prices. One way to predict the value of stock prices using an econophysics approach is predict using machine learning. One type of machine learning that can be used for technical analysis is the Gated Recurrent Unit (GRU) and L...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/65582 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Econophysics is a field that we can use in predicting the value of stock prices. One way to predict the value of stock prices using an econophysics approach is predict using machine learning. One type of machine learning that can be used for technical analysis is the Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) which are part of the recurrent neural network. In a recurrent neural network, some hyperparameters can be set so that the method can learn optimally. This study aims to determine the value and type of hyperparameters that provide the most optimal results for the GRU and LSTM methods and also compare the results of stock price predictions for the GRU and LSTM methods. The study was conducted by varying the epoch and learning rate, and batch size using data from 3 Indonesian companies with different trends, namely BBRI, UNVR, and PGAS. The results of stock price predictions will be analyzed by looking at the values of the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the Coefficient of Determination (R2). From the results of this experiment, it was concluded that the epoch values that gave the most optimal results were 100 and 200 for the GRU and LSTM methods. Meanwhile, the learning rate that gives the most optimal results is 0.0004 for the GRU method and 0.0005 for the LSTM method. Then the batch size value that gives the most optimal results for both the GRU and LSTM methods is 32 The most influential hyperparameters on the prediction results are batch size, followed by epoch, and last learning rate. From the results of this experiment, it is also concluded that the GRU method has better accuracy results and converges faster to achieve the smallest loss value when predicting stock price values. |
|---|