OPTIMIZATION OF BI-GRU ON HYPERPARAMETER VARIATION AND ACTIVATION FUNCTION FOR K-POP AGENCY STOCK PREDICTION

• Main Author: Gularni Purnawulan, Dascha
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78256
• Institution: Institut Teknologi Bandung

Econophysics is a science that applies concepts and methods originally developed by physicists to solve economic problems. One of the applications of econophysics is modeling stock price movements. This research utilizes the Bidirectional Gated Recurrent Unit (Bi-GRU) model inspired by physics concepts, such as the law of conservation of momentum and the law of conservation of energy. In the Bi-GRU model, there are hyperparameters and activation functions that can be adjusted to produce more optimal model. The purpose of this research is to determine the effects of varying the type and value of hyperparameters, as well as determining the activation function that provides the best Bi-GRU model performance in predicting the stock price of k-pop entertainment agencies. The research was conducted by varying three hyperparameters, namely epoch, learning rate, and batch size. After that, modeling is done by varying three activation functions, namely sigmoid function, tanh function, and ReLu function. In the research results, it was found that large epoch value causes the model to be overfitting, while small epoch value causes the model to be underfitting. Large learning rate value causes the model to overshoot, while small learning rate value causes the model to be too slow towards the optimal solution. Large batch size value causes the computation process to be slow, while small batch size value causes the model to be overfitting.