Performance Analysis of Bidirectional-LSTM and GRU Models for Financial Time-Series Forecasting

This paper explores the deep learning-based models for predicting stock price movements in the SET50 index. By leveraging advanced architectures, including Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Gated Recurrent Unit (GRU), the study investigates the models’ ability to handle the sequential and non-linear nature of financial time-series data. Preprocessing steps, model evaluation, and performance metrics such as Mean Squared Error (MSE) and directional accuracy are discussed. The findings demonstrate that Bi-LSTM outperforms compared with other models in terms of accuracy, while GRU offers a more computationally efficient solution. This work contributes to improving stock market forecasting methodologies and provides practical insights for financial applications.