<![CDATA[This study aims to develop and evaluate a prediction model for the movement of the Composite Stock Price Index (JCI) using the Long Short-Term Memory (LSTM) method as one of the effective deep learning approaches in capturing nonlinear time series patterns. This study tested three model approaches, namely the Univariate model which only uses historical JCI data, the Multivariate All Feature model which integrates all external variables, and the Multivariate Selected Feature model which uses selected external variables. External variables considered include world gold prices, world oil prices, rupiah exchange rates against the United States dollar, and international stock indices that affect the Indonesian capital market. The model performance evaluation was carried out using the Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) indicators. The results showed that the Multivariate All Feature model provided the best performance with a MAPE value of 0.76, an RMSE of 66.72, and an MAE of 51.58, which was consistently lower than the other two models. Significance tests using ANOVA and Tukey HSD confirmed significant performance differences between models. These findings indicate that the integration of external variables is able to significantly increase the accuracy of JCI predictions. This research is expected to be a reference for investors, capital market analysts, and researchers in the development of artificial intelligence-based decision support systems in the Indonesian stock market.]]>
Copyrights © 2025
