Implementasi Model BiLSTM-Attention untuk Prediksi Nilai IHSG Berdasarkan Data Historis OHLCV

Abstract

The Composite Stock Price Index (IHSG) reflects the performance of the Indonesian capital market, but predicting it is challenging due to high volatility and the influence of various external factors. This study aims to develop and evaluate a deep learning-based predictive model using a Bidirectional Long Short-Term Memory (BiLSTM) architecture combined with an Attention Mechanism to predict the IHSG value based on historical numerical data (OHLCV). This method was chosen for its ability to recognize bidirectional sequential patterns and highlight the most relevant historical information in the prediction process. The research was conducted quantitatively using an experimental approach, and model evaluation was performed using regression metrics such as R², RMSE, MAE, and MAPE. The results obtained showed excellent predictive performance with an R² of 0.9485, MAPE of 0.63%, RMSE of 59.47, and MAE of 45.12. Additionally, attention weight analysis revealed that the model focuses more on the last two days within the prediction time window, indicating that recent information significantly influences IHSG movements. These findings suggest that the BiLSTM-Attention approach is effective in capturing stock market dynamics and has the potential to serve as a strategic tool for data-driven investment decision-making.